Vol.

3
Sept 2024

Innovate With AI:

The Future Enterprise
Driving industry success
with artificial intelligence
and MongoDB Atlas

Atlas for Industries

A look at how AI and
MongoDB are creating
value across industries

AI and Application
Modernization
Modernizing apps faster for
a better user experience

1
 Table of Contents

3 Editor’s Introduction 68-71 Assessing Business Loan Risks

72-73 Emerging use cases for AI in payments
4 AI and the Developer Data Platform
74-76 Regulatory Compliance and Code
7-28 Manufacturing and Motion Change Assistance
8 AI Investments 77 Other Notable Use Cases
9-15 Inventory Management 77 Contact Information
16-24 Predictive Maintenance 78-93 Insurance
25-27 Knowledge Management
28 Other Notable Use Cases
79-84 Architecting with AI
28 Contact Information
84-86 Underwriting & Risk Management
86-87 Claim Processing
29-37 Telecommunications and Media 88 Customer Experience
30-31 AI Investments 89-92 Domain-Driven AI & Solution Library
32 Service Assurance 93 Other Notable Use Cases
33 Fraud Detection and Prevention 93 Contact Information
34-35 Content Discovery
94-110 Healthcare and Life Sciences
36 Search Generative Experiences (SGE)
37 Other Notable Use Cases 95-96 Potential of Generative AI
37 Contact Information 97-98 Patient Experience and Engagement
99-100 Enhanced Clinical Decision Making
38-51 Retail 101-103 Enhancing Medical Imaging with
39-40 AI Investments Generative AI
41-43 AI-Augmented Search and Vector 104-105 Revolutionizing Clinical Study Reports
Search 106-109 Accelerating Drug Discovery
43-46 Personalized Marketing & Content 110 Other Notable Use Cases
Generation 110 Contact Information
46-48 Demand Forecasting & Predictive
Analytics 111-121 AI Leaders in Action
49-51 Conversational Chatbots 112-116 Ben Ellencweig from McKinsey
51 Contact Information 117-121 Asaf Somekh from Iguazio

52-77 Financial Services 122-132 AI and Application Modernization

Programs
53-54 AI Investments and Challenges
55-57 Relationship Management 133-225 AI Partnerships
Support with Chat Bots
134-137 Hyperscaler cloud services
58-59 Risk and Fraud Prevention
138-143 System Integrators
60-61 Financial Document Search and
144-190 AI With SaaS
Summarization
191-225 Component-Based AI
61-63 ESG Analysis
64-67 Transforming Credit Scoring with AI 226-227 Conclusion and Next Steps
Editor’s
Introduction

I am delighted to present our third edition of the art of possible at the

the dynamic intersection of AI, innovation, and industry solutions. This
e-book serves as a beacon, guiding readers through the intricate
landscape of AI solutions across not only the industry your organization
sits in, but also provides insights into how other industries are innovating
with AI. Along the way, we will explore the top use cases across the six
core industries that are infused with MongoDB Atlas AI capabilities. I
strongly suggest to browse use cases from other industries - often they
are applicable in a changed context as well.

Why read it, you ask? Because within these pages lie invaluable insights
into the critical role of AI. Understanding its significance and harnessing
its power is paramount for businesses striving for success. You can also
delve into our partner section highlighting organizations that have built
AI solutions using MongoDB and our MongoDB AI Application Program
(MAAP). Whether a SaaS end-to-end solution you can implement, or
component-based solution you can plug in, there is something here for
you.

We invite you to take advantage of our innovation workshops. These

ideation sessions are available to you at your location in your
environment. You will meet MongoDB industry experts, our partners and
discuss the art of the possible based on your requirements and needs.
Additionally check out our MongoDB AI Applications Program on the
MongoDB website designed to help organizations rapidly build and
deploy modern applications enriched with AI technology at enterprise
scale.

Boris Bialek: Vice President and Field CTO, Industry at MongoDB

 AI and the Developer
Data Platform
AI is quickly becoming a universal tool that fits deploying machine learning models or
in every industry’s toolbox. Soon after early integrating cutting-edge generative AI into
machine learning and AI predictive capabilities your applications, MongoDB Atlas is an
harnessed the power of big data to give indispensable component of your technology
enterprises deeper business analytics at stack. From inception to deployment,
eye-popping speed, new advances in MongoDB Atlas ensures that your applications
generative machine learning applications such are grounded in accurate operational data
as OpenAI and Hugging Face opened while meeting the demands of scalability,
possibilities for generating and analyzing text security, and performance expected by users.
data. Today, generative AI (gen AI)
applications go beyond text data, creating MongoDB has seen widespread adoption for
hyper-personalized experiences. traditional AI use cases. Continental selected
MongoDB for the feature engineering platform
While implementing AI technology can be in its Vision Zero autonomous driving initiative.
risky, complex, and time-consuming, the Both Bosch and Telefonica use MongoDB in
potential for benefits such as higher profits, their AI-enhanced IoT platforms. Kronos
faster innovation, and lower costs are driving trades billions of dollars' worth of
industries toward an AI-powered future. cryptocurrency daily using machine learning
MongoDB Atlas, the ground-breaking (ML) models built with data from MongoDB.
developer data platform, integrates Iguazio uses MongoDB as the persistence
operational, analytical, and gen AI layer for its data science and MLOps platform,
data services, simplifying the development of while H2O.ai and Featureform support
intelligent applications. Whether you're MongoDB as feature stores in their platforms.

At the heart of MongoDB Atlas lies its flexible

Flexible document data model and developer-friendly

Data Model
query API. Together, they empower developers
to accelerate innovation, gain a competitive
edge, and seize new market opportunities
presented by gen AI. Documents, which align
seamlessly with code objects, offer an intuitive
and adaptable way to manage data of any
structure. Unlike traditional tabular data
models, documents afford the flexibility to
accommodate diverse data types and
application features, facilitating data
rationalization and utilization in ways
previously
US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. unattainable.
All rights reserved.
 Rapid
Querying

Paired with the document model, the search, enriching the user experience and
MongoDB Query API provides developers with expanding application capabilities. MongoDB
a unified and consistent approach to data Atlas stands apart by offering a
manipulation across various data services. comprehensive suite of query functionality
From basic CRUD operations to complex within a single, unified experience. This
analytics and stream processing, the MongoDB eliminates the need for developers to manually
Query API offers developers the flexibility to integrate query results from multiple
query and process data according to the databases, reducing complexity, errors, costs,
application's requirements. In the realm of gen and latency. Moreover, it maintains a compact
AI, this flexibility enables developers to define and agile technology footprint, enabling
additional filters on vector-based queries, such developers to focus on building end-user
as combining metadata, aggregations, and functionality with greater ease and efficiency.
geospatial

Across retail, manufacturing,

The Rise of telecommunications, and insurance industries,

companies are often falling short on their

Real-Time ambitions to build data-driven operations as

they struggle to perfect real-time analytics

Analytics using real-time data.

& Dynamic With MongoDB Atlas App Services, these

industries are able to reinvent pricing

Pricing strategies to reflect market fluctuations,

demand surges, or coverage changes. It’s key
to recognizing the importance of transforming
raw data into a more usable structure and
understanding the benefits of serverless
functions and triggers, which can
automatically respond to changes in data and
execute predefined actions with a dedicated
server.

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 Vectors, Unstructured Data, and
MongoDB Atlas Vector Search

To feed AI models with proprietary data, there to with structured business data. Considering
is a need to create vector embeddings. Data in that more than 80% of data is unstructured,
any digital format and of any structure – i.e., it’s easy to appreciate how transformational
text, video, audio, images, code, tables – can vector search combined with gen AI really is.
be transformed into a vector by processing it
with a suitable vector embedding model. This Once data has been transformed into vector
incredible transformation turns data that was embeddings, it is persisted and indexed in a
previously unstructured and, therefore, vector store such as MongoDB Atlas Vector
completely opaque to a computer into data Search. To retrieve similar vectors, the store is
that contains meaning and structure inferred queried with an Approximate Nearest
and represented via these embeddings. Now Neighbor (ANN) algorithm to perform a K
users can search and compute unstructured Nearest Neighbor (KNN) search using an
data in the same way they’ve always been able algorithm such as 'Hierarchical Navigable
Small Worlds' (HNSW).

Figure 1: Data is transformed from unstructured internal sources to actionable, impactful insights

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 Atlas for Industries

Manufacturing
and Motion
The integration of AI
within the manufacturing
and automotive industry
has transformed the
conventional value chain,
presenting a spectrum
of opportunities.
AI investments in the manufacturing industry

AI in manufacturing is a game-changer. It has an optimal combination of several

the potential to transform performance across characteristics, that make them an ideal place
the breadth and depth of manufacturing to start:
operations. Companies are in a race to ● Clear business value/benefits
embrace AI, as these technologies are critical ● Relative ease of implementation
enablers of the Fifth Industrial Revolution (also ● Availability of data
known as Industry 5.0). Artificial intelligence in
manufacturing is bringing factories into the Smart manufacturing use cases are
future and will ultimately empower the revolutionizing many organizations, and a key
manufacturing market to continue to be the driver of this is the incorporation of artificial
backbone of the global economy. intelligence into manufacturing processes.
Many firms have embarked on significant
Industry-wide, manufacturers are facing a digital transformation journeys in the past two
range of challenges that make it difficult to years with the goal of improving efficiency and
speed production while still providing resilience. However, a concerning gap exists
high-value and high-quality products to their between tech adoption and return on
customers. All the while, companies need to investment. While 89% of organizations have
implement a digital infrastructure that begun digital and AI transformations, only 31%
positions them to fully embrace the skills and have seen the expected revenue lift, and only
knowledge of their best assets — people. 25% have realized the expected cost savings
(McKinsey). In some cases, situations have
The manufacturing industry today relies on even worsened.
automation just as much as people. But the
factory of the future, which is a marriage of This highlights the importance of not just
physical and digital capabilities, requires more: implementing new technologies, but
real-time data, connectivity and AI technology implementing them strategically. In other
at the forefront. In fact, more than 80% of words, simply deploying AI isn't a guaranteed
C-suite executives believe they must leverage path to success. Manufacturers need to
AI to achieve their growth objectives. carefully consider how AI can address their
Customer requirements for delivering on-time specific challenges, and then integrate it into
and on-budget product are of the utmost existing processes effectively.
importance, and efficiency is a goal in
everything. This chapter unpacks how major players in
the manufacturing industry are leveraging
AI’s ability to drive impact in this regard is real. AI to improve operations, deliver better
According to a study of Capgemini, three use outcomes for customers, and realize
cases stand out in terms of their suitability for innovation. It delves into three high impact
kickstarting a manufacturer’s AI journey: value drivers and AI use cases: Predictive
demand planning, intelligent maintenance and Maintenance, Inventory Management and
product quality control. These use cases have Knowledge Management.
The path to success

Successful organizations exhibit common ● Continuous data enrichment and

traits across five key areas: accessibility: High-quality data, readily
available and usable across the
● Identifying high-impact value organization, is essential for the success
drivers and AI use cases: Efforts of AI initiatives.
should be concentrated on domains ● Empowering talent and fostering
where artificial intelligence yields development: By equipping their
maximal utility rather than employing it workforce with training and resources,
arbitrarily. organizations can empower them to
● Aligning AI strategy with data leverage AI effectively.
strategy: Organizations must establish ● Enabling scalable AI adoption:
a strong data foundation with a data Building a strong and scalable
strategy that directly supports their AI infrastructure is key to unlocking the full
goals. potential of AI by enabling its smooth
and ongoing integration across the
organization.

Inventory Management
and Optimization

Current State and Challenges balance between inventory levels to

Modern manufacturing supply chains are maximize profitability and competitive
complex systems, interconnected across the advantage in the market. Effective inventory
globe. Efficient supply chains are able to management mitigates the risk of ‘bullwhip
control operational costs and ensure on-time effect’, where sudden demands can disrupt
delivery to their customers. Inventory the supply chain costs and performance.
optimization and management is a key
component in achieving these goals. While Key components of supply chain
maintaining higher inventory levels allows for management include procurement and
suppliers to deal with unexpected sourcing, manufacturing and production,
fluctuations in demand, they come with distribution, logistics and retail.
higher inventory holding costs that may be Technological advancements including IoT
passed on to customers. Thus, every player and AI (including Gen AI) are being
in the supply chain is motivated to strike a integrated into SCM to improve
transparency, efficiency and adaptability of between OEMs, tier1-n suppliers and
the supply chain, allowing for real time customers, always aiming for reduced costs,
monitoring, predictive analytics, and enhanced quicker production and response times and
decision-making capabilities. heightened customer satisfaction, all of which
will result in stronger market position.
technology-driven ecosystem that requires
collaboration throughout the supply chain

Figure 2: A representation of an automotive supply chain

Inventory management is essential for right inventory in the right place at the right
achieving the goals of efficient supply chains, time to minimize system costs and meet
controlling costs, and delivering to customers customer needs.
with minimal delays. Inventory management
is primarily concerned with the planning and Usually, companies do supply chain planning
controlling an industry’s inventory. It includes at several levels, namely strategic, tactical, and
business processes such as estimating operational. Each level differs in its objectives,
material requirements at various points in the planning horizon, and level of detail. Strategic
supply chain, determining necessary material’s and tactical planning are crucial to successful
amount, ordering frequency, and safety stock supply chain management. The so-called
levels. It also includes inventory visibility, Pareto's law could be applicable here as 20%
inventory forecasting, lead time management, of efforts in strategic and tactical planning
inventory shipping costs, inventory valuation, brings 80% of the total effect. At the strategic
forecasting future inventory prices, available level, the leadership team makes high-level
physical space, quality management, returns decisions that affect the entire organization.
and defective goods, and demand forecasting. Scenario planning is done at this level. The
It plays a very important role in reducing analysts go through scores of internal and
overall costs and rapid response objectives. external data including global news, political
Effective inventory management requires the developments, think tank studies and scientific
literature to pinpoint the most strategic integration challenge. The spread of data
concerns and trends that the organization across multiple systems and locations
needs to focus on. The team can then use (on-prem systems, cloud regions etc.) can lead
these outputs to develop a set of draft to data silos and hinder visibility into overall
scenarios for consideration. inventory levels and movements. Finally, when
there is sparse historical inventory data
This tedious process comes with its own set of available, then traditional ML models may
challenges. Predicting long-term demand, suffer in accuracy.
market trends and economic conditions is
challenging because of the long-term horizon. Generative AI and IoT technologies hold
This long planning horizon increases the potential to address some of these challenges.
uncertainty in predicting demand, as market Generative AI in particular can assist in
conditions, consumer preferences, and scenario planning by generating various
technological advancements can change potential outcomes based on a wide range of
significantly over time. data, allowing the organization to prepare
better for an uncertain future.
At the tactical and operational level, for
manufacturers, to manage and optimize How AI and MongoDB Help
inventory levels, the first step is to maintain an We will start with scenario planning to
accurate and real-time view of inventory levels generate hypothetical situations which could
across multiple plants, warehouses, and affect inventory requirements, supply chain
suppliers. This is absolutely essential as performance and overall operations. Effective
without having visibility on the current scenario planning helps companies plan for
inventory levels, it is impossible to optimize. optimal inventory levels.
The second step is to reduce inventory carrying
costs while still ensuring that the required parts After scenario planning, we will look into
and materials are available to ship out when inventory classification use cases where AI can
needed. Finally, the data from multiple be used to categorize inventory based on
customers needs to be aggregated and factors such as demand variability, lead times
analyzed despite being in different formats, and criticality.
each with its own unique lead times and order
quantities. Scenario planning
A scenario planning process has two stages
However, efficient inventory management for
manufacturers presents complex data ● Scenario generation
challenges too, primarily in forecasting ● Strategy generation
demand accurately and optimizing stock
In scenario generation, a generative AI
levels. One issue routinely faced is the
application can look at a vast amount of data
variability and unpredictability of customer
including internal and external business data,
demand patterns, making it difficult to
competition data, political news and events
precisely anticipate inventory needs. Managing
and social media news, find the correlation
diverse data streams from sales records,
between each piece of unstructured
production schedules, supplier information,
information and then rank these areas of
and market trends poses a considerable data
concern in terms of their estimated
 significance. It is important to utilize general
knowledge that a Large Language Model
processes as well as internal company data in a
Retrieval Augmented Generation (RAG) model
to avoid hallucinations.

Figure 3: A Scenario Generation RAG App with MongoDB

The figure above shows a scenario generation factors due to certain reasons, including but not
application data flow. This retrieval-augmented limited to the political landscape and global
generation (RAG) system consists of four parts. supply chain disruptions etc.
First, an AI data store aggregates and
operationalizes structured and unstructured MongoDB Atlas streamlines RAG
data. In our case, the majority of the data is in implementation as it handles everything under
unstructured format such as news sources, the hood. App data, metadata, context data,
social media and competitive information and vectors are all stored in the same place. As
typically stored as unstructured PDF files. The the app evolves, the document model is
PDF files are vectorized and stored in MongoDB inherently flexible and ideal for storing
Atlas. Atlas Vector Search is then utilized to structured and unstructured data. You can add
perform semantic searches and to find data to the same collections inside the
meaningful context from the PDF embeddings. database as needed, vectorize it, and store the
vectors alongside it.
Atlas Vector Search can be triggered using an
AI application, connecting to MongoDB Atlas to Once the data is stored, vector search
retrieve the right context, which is then fed into capabilities are provided right out of the box,
the large language model to answer questions and search operations can be optimized using
like "Can you generate hypothetical trends that dedicated search nodes.
can affect our firm?" The response might
suggest customer behavior shifts or economic

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
With MongoDB Atlas, it's just one query in one Once the demand is known, AI can play a
technology, one language, and one pivotal role in stock optimization. By analyzing
infrastructure to manage and scale, with no historical sales data, market trends,
data duplication, ultimately leading to a lower manufacturers can determine the most
total cost of ownership and a unified developer efficient stock levels. AI systems can also place
experience. orders automatically based on predicted
demand and targeted stock levels. This
The same application can then categorize the automation not just saves time but also
trends in terms of their probability of reduces human error. Finally, AI utility can be
occurrence and impact. Next, a strategy can extended to supplier selection and relationship
be created automatically, which provides management. By analyzing supplier
guidance on what steps to take in response to performance data, AI based software can
the trends and scenarios generated. The assist in choosing the right suppliers who will
strategy can also contain information about meet the company’s quality, delivery and cost
the financial implications and risks associated requirements.
with the response, for example any significant
initial investment in hiring more AI scientists to MongoDB Atlas provides a flexible, scalable,
develop RAG applications for inventory and highly available developer data platform
management. for managing inventory data. The document
data model can handle complex inventory
For scenario planning, generative AI can be structures and hierarchies, making it easy to
used to generate and evaluate strategies. manage inventory across multiple plants and
However, it is important to provide enough suppliers.
context to the LLM so that it does not
hallucinate. MongoDB Atlas Vector Search is At the warehouse, the inventory can be
key to creating a RAG application. Additionally, scanned using a mobile device. This data can
these AI-generated strategies should be be persisted in a MongoDB collection. Once
viewed as initial concepts for further data is in Atlas, it can serve as the central
exploration rather than as final solutions to be repository for all inventory-related data which
adopted without additional analysis. includes stock, supplier, and customer
Generative AI should complement human information, bill of materials and production
efforts by supporting the identification, line data. This repository becomes the source
evaluation and timely execution of appropriate of data for the inventory management AI
strategies. applications. This approach removes data silos
and improves visibility into overall inventory
Inventory Optimization levels and movements.
One of the most significant applications of AI
in inventory management is in demand However, the challenge of poor or sparse data
forecasting. AI algorithms can be used to at the source systems may still remain. To solve
analyze complex datasets to predict future this, manufacturers can take advantage of
demand of products or parts. Improvement in generative AI and Atlas Vector Search to
demand forecasting accuracy is crucial for implement a Retrieval Augmented Generation
maintaining optimal inventory levels. AI-based (RAG) architecture to generate synthetic data
time series forecasting can assist in adapting whenever needed. They can take multimodal
to rapid changes in customer demand. content such as product descriptions and
specifications, customer feedback and reviews RAG. They can easily categorize products
and inventory notes, vectorize them and store based on their seasonal attributes, cluster
vector embeddings alongside the operational products with similar seasonal demand
data in MongoDB Atlas. This allows them to patterns and provide context to the foundation
supercharge their inventory optimization using model to improve the accuracy of synthetic
inventory data generation.

Figure 4: Enhancing Inventory Data with Vector Embeddings

The figure below shows a reference customer behaviors. This approach not only
architecture of generative AI+AI enabled fills the gap of missing historical data for new
demand forecasting with MongoDB Atlas. The products but also provides a robust foundation
accurate demand forecasting results will help for demand forecasting enabling
in stocking up on right inventory levels. For manufacturers to optimize their inventory
new products, the historical sales data is not levels.
available. Generative AI models can create
realistic and diverse synthetic data by learning
patterns from existing datasets of similar
products. This synthetic data can mimic the
sales trends and seasonality that new products
Solution demo
Discover how to building an
might experience. Atlas Vector search can find
event-driven inventory management
similar products attributes and feed that
system.
context into the generative AI model. By
finding semantics in similar products, Atlas
vector search can help refine the synthetic
data generation, ensuring that it closely
reflects potential market conditions and
Figure 5: Gen AI enabled demand forecasting with MongoDB Atlas

Ceto is revolutionizing maritime MongoDB's architecture provided Ceto with

operations with MongoDB time several key features that are crucial for their
operations. Scalability was essential for
series
managing the increasing data volumes
Ceto is on a mission to bring the maritime generated by their expanding fleet. Time
industry into the digital age—and to Series Collections offered advanced data
transform maritime operations into a model compression capabilities, crucial for
of efficiency and sustainability. managing the large volumes of data
generated daily.
To make its mission a reality, Ceto
partnered with MongoDB, leveraging its “MongoDBʼs Time Series collections
robust data handling capabilities to
have revolutionized how we
integrate AI with real-time data collected
from thousands of sensors across its
manage and utilize data from our
customers’ fleets. This allows Ceto to predict fleet. The ability to process and
and preempt potential failures, streamline analyze data in real-time has
operations, and manage risks proactively.
significantly enhanced our
This shift not only enhances safety and
reliability but also propels maritime logistics predictive maintenance
into a new era of technological capabilities.”
advancement, making Ceto a
Learn more
transformative force in global commerce.
Ben Harrison
CTO, Ceto
Predictive Maintenance

A well-defined maintenance strategy can be a Third, based on the prediction, the system can
game-changer for any organization, driving generate work orders, schedule maintenance
significant revenue and cost savings. Here's activities, and even provide guidance to
how it works: maintenance personnel. This ensures that
maintenance is performed only when
First, identify the equipment that is most necessary, optimizing resource allocation.
crucial for your operations. Downtime for this
equipment can lead to bottlenecks, halting This series of activities delivers tangible
production. benefits. Costs are reduced through saved
labor hours and extended machine lifespan.
Second, equip these critical assets with Additionally, revenue increases as your
sensors to enable condition monitoring. This machines operate at optimal performance
allows you to monitor the health of the levels.
equipment in real time, identifying potential
issues before they escalate into catastrophic
failures.

Figure 6: Steps required for an optimal maintenance strategy

 Optimal Maintenance Strategy which allows for proactive maintenance
and minimizes downtime. Predictive
An optimal maintenance strategy isn't a maintenance requires significant
one-size-fits-all solution. It's about choosing upfront investment in sensors and data
the right blend of approaches based on your analysis tools.
specific equipment and operational needs.
Predictive maintenance uses data analysis to
In today's processes, we see a spectrum of identify problems in machines before they fail.
maintenance approaches. These methods This allows organizations to schedule
range from highly complex and expensive at maintenance at the optimal time, maximizing
one end to simpler, more affordable options on machine reliability and efficiency.
the other:
Here's how predictive maintenance can benefit
● Reactive maintenance: This is the manufacturing operations, according to
most basic approach in which Deloitte:
maintenance is performed after a
machine fails. While simple, it can lead 3-5% Reduction in new equipment costs
to unexpected downtime and higher
repair costs. 5-20% increase in labor productivity
● Preventive maintenance: This is a
proactive approach that involves 15-20% reduction in facility downtime
scheduling maintenance tasks based
on predetermined time intervals or 10-30% reduction in inventory levels
usage metrics. This helps prevent
breakdowns but can be inefficient as 5-20% reduction in carrying costs
machine conditions can vary.
Predictive maintenance is constantly evolving.
Thresholds for these tasks may need to
We've moved beyond basic threshold-based
be adjusted due to factors like aging
monitoring to advanced techniques like
equipment, changes in processes, or
machine learning (ML) models. These models
different materials being used.
can not only predict failures but also diagnose
● Condition-based maintenance
the root cause, allowing for targeted repairs.
(CBM): This approach continuously
monitors the health of the machine
The latest trend in predictive maintenance is
through sensors and data analysis.
automated strategy creation. This involves
Maintenance is then triggered based
using AI to not only predict equipment
on the actual condition of the
breakdowns but also to generate repair plans,
equipment rather than a set schedule.
ensuring the right fixes are made at the right
This is more efficient than preventive
time.
maintenance as it avoids unnecessary
maintenance. Threshold-based alerting Automated strategy creation requires
systems are often used with CBM. substantial investment in R&D, along with deep
● Predictive maintenance: This is the industry knowledge, access to relevant data,
most advanced approach, using data and practical operational experience. The
analytics to predict potential question is, can generative AI help?
equipment failures before they occur,
US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 Current State and Challenges

The answer is yes, generative AI can help. But

there are challenges at each stage of
implementation that organizations must
consider. Each stage involves a key question
and associated challenges, highlighting the
steps and issues faced in predictive
maintenance and machinery upkeep.

Figure 7: Different challenges seen at each stage of the predictive maintenance deployment

Now, let's envision a factory containing both maintenance managers and leaders and
automated and manual machines. Let's call it conducting quantitative analyses, a rather
Gear Transmission Systems Ltd, whose primary manual process.
output is gearboxes.
Once you’ve identified the machines, the next
Within this factory, we have an array of step is to install sensors and train the machine
equipment: cutting machines, milling learning model. However, two major challenges
machines, measurement devices, and more. As arise. First, you lack sufficient "run to failure"
its general manager, you are tasked with data to effectively train the model. Secondly,
managing the budget allocated for machine health deteriorates over time, leading
maintenance processes and improving to evolving failure modes with the age of the
strategies. One pressing question you must machine.
address is: which machines should take priority
for the predictive maintenance projects, and Assuming you manage to overcome these
why? This involves consulting with the hurdles, the next phase involves maintenance

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 scheduling and execution. You’re faced with a ● Prioritization of
myriad of data silos, including inventory data machines/criticality analysis: At this
and resource availability data, which need to stage, we require both structured and
be integrated to formulate a comprehensive unstructured data. We need previous
repair plan. Furthermore, it's essential to machine failure data as well as expert
ensure that operators are adept at addressing analysis/opinion on which machines to
minor machine issues to reduce reliance on prioritize for predictive maintenance
external experts. While complex issues may and why.
still require OEM or SI support, internal ● Failure prediction: This stage involves
troubleshooting capabilities are invaluable. structured data such as sensor data
Therefore, developing easy-to-follow and maintenance logs to identify the
documentation tailored to the skill levels of our root cause of imminent failure.
staff is important. ● Maintenance plan and guidance
generation: In both of these stages,
As the Figure below shows, different data is we deal with both structured and
required for solving above listed challenges: unstructured data. The objective is to
combine this data to generate an
optimal repair plan and operator
guidance.

Figure 8: Different data requirements at each stage

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 How AI and MongoDB Help and to provide the right context to the large
language model, ensuring it generates
MongoDB Atlas is the only multi-cloud responses based on factory data without
developer data platform designed to hallucinating. Such a system can result in
accelerate and simplify how developers work positive business outcomes. Here’s how the
with data. Using MongoDB Atlas, developers system can look with MongoDB Atlas as the AI
can power end-to-end value chain data store:
optimization with AI/ML, advanced analytics,
and real-time data processing for innovative
mobile, edge, and IoT applications.

Stage 1: Machine prioritization

Current machine prioritization for predictive

maintenance relies heavily on manual analysis.
Factory personnel gather historical and current
machine data on utilization losses due to
breakdowns. This data is then reviewed
alongside the experience of maintenance
managers and leaders. Based on this
combined analysis, a roadmap for the
Figure 9: Generative AI-powered machine
predictive maintenance project is
prioritization
recommended, highlighting which machines
should be prioritized.
This retrieval-augmented generation (RAG)
However, this approach has limitations. A system consists of four parts. First, an AI data
reliance on manual analysis can be store aggregates and operationalizes
time-consuming and may not always capture structured and unstructured data. In the Figure
the full picture of the maintenance project due above, machine breakdown history and
to the limited use of quantitative data sources. operational parameters are represented as
Additionally, inconsistencies in interpretation structured data, while expert interviews are
can lead to an overdependence on institutional stored as unstructured PDF files. The PDF files
knowledge, which in turn can result in false are vectorized and stored in MongoDB Atlas.
analyses that impact the project's return on Atlas Vector Search is then utilized to perform
investment (ROI). semantic searches and to find meaningful
context from the PDF embeddings.
But with the arrival of generative AI, things
have changed. A generative AI-based machine Atlas Vector Search can be triggered using an
prioritization tool can be created to reduce the AI application, connecting to MongoDB Atlas
time manufacturing experts spend on manual to retrieve the right context, which is then fed
analysis, and to decrease the risk of poor into the large language model to answer
investments. To leverage AI, experts need a questions like "Which machine should I
data store capable of storing and prioritize and why?" The response might
operationalizing both structured and suggest prioritizing Machine M001 or M002
unstructured data. Having such a data store due to certain reasons, including but not
US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
will allow them to perform semantic searches limited to the criticality of the machine, high
maintenance cost, etc.
 Stage 2: Failure prediction operational data, real-time processing
capabilities, integrated monitoring and
Now that we've discussed prioritizing alerting, and seamless compatibility with
equipment, let's move on to failure prediction. machine learning tools, MongoDB Atlas
MongoDB Atlas provides all the necessary enables organizations to optimize machine
building blocks or tools to implement failure performance and minimize downtime.
prediction. By providing a unified view of

Figure 10: Smart milling machine uses real-time data to predict failures

As seen in the Figure above, we have our topic. This allows data transformation and
prioritized machine, which is a milling machine enrichment before it even lands in the
with attached sensors that collect data such as database.
air temperature, rotational speed, torque, and
tool wear. This data will be processed through Once the data is in MongoDB, another
Atlas Stream Processing, enabling the application can react to sensor values and run
processing of streams of complex data using a trained model designed to predict failures.
the same data model and Query API used in The model results can be stored back into
Atlas databases. Atlas Stream Processing Atlas (between steps 4 and 5/Inference in the
enables developers to build aggregation Figure above). These results can then be
pipelines to continuously operate on streaming visualized using Atlas Charts. Finally, Atlas
data without the delays inherent to batch Triggers and Functions can be used to push
processing. Results can be continuously notifications to on-site users. This establishes
published to MongoDB Atlas or to a Kafka an end-to-end system for failure prediction.

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 Stage 3: Repair plan generators schedule, and resource availability information.
In this case, both structured and unstructured
Having identified the nature of the equipment data are involved. The repair instructions will
failures, the implementation of a come from the machine manual. For this
comprehensive repair strategy becomes process, MongoDB Atlas acts as the
paramount. First, we have to generate a operational data layer, seamlessly integrating
maintenance work order. This order should structured and unstructured data.
include repair instructions, spare parts needed,

Figure 11: MongoDB Atlas as the operational data layer for structured and unstructured data

The Figure above shows the process of work old work orders. What sets this approach apart
order generation using generative AI. First, we is the ability to use the same MongoDB
must extract chunks of information from a database to store structured data such as
milling machine's operating and repair manual, failure types, spare parts inventory, and
as well as from old work orders stored as PDF resource information. By employing the
files, and convert them into vectors. These aggregation framework to extract relevant
embeddings are then stored in MongoDB Atlas. information from structured data and vector
MongoDB's versatility allows for the storage of search to glean insights from vectors, the LLM
both structured and unstructured data within within the work order generator application
the same database. Leveraging Atlas Vector gains contextual understanding.
Search and Aggregation pipelines, we can
integrate this data to feed into a large This application seamlessly utilizes the LLM to
language model (LLM) powering a work order generate work orders with the correct
generator application. The LLM analyzes the template, filling in inventory and resource
data to generate the appropriate work order details through aggregations, and ultimately
and template, drawing from past examples. It creating repair plans based on machine
populates inventory and resource details using manuals. This application can run inside a
aggregation
US 866-237-8815 techniques
| INTL +1-650-440-4474 and structured
| [email protected] 2023 MongoDB, Inc.central
| © data. maintenance
All rights reserved. management system.
Finally, it generates a repair plan similar to the
 Stage 4: Maintenance guidance generation

So we come to the last step: How can we use

gen AI to enhance the operator or technician
guidance to maintain the machine?

Figure 12: Using the RAG Approach for Operator Work Instructions

Let's walk through an example scenario here. then we can publish notifications down to the
The repair plan was generated in the last step. technician's application via Atlas Triggers and
Now, the computerized maintenance Functions.
management system (CMMS) has found some
service notes uploaded to the platform by In summary, we are essentially integrating AI
another technician, but they're written in and gen AI apps to implement an end-to-end
another language, let's say Spanish. We can predictive maintenance strategy (shown in the
use the RAG architecture again to intelligently Figure on the next page).
merge these service notes with the repair
instructions generated in the previous step.

We first need to extract text from the PDF,

translate it into English since our other data is
in English, and then provide the service notes
as well as the repair plan to the LLM as
context. So, we have the original plan from the
previous steps, and we combine it integrally
using the LLM with the service notes obtained
in this step. Note that we're not performing
vector search here. Once the plan is updated,

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 Figure 13: Model chaining with unified data store

Our input consisted of a combination of

structured and unstructured data. We
leveraged the various services offered by the Solution Demo
MongoDB Atlas developer data platform,
Find out how AI is being used in
including Atlas Vector Search, Atlas Stream
renewable energy by leveraging
Processing, and, of course, the MongoDB
database and aggregation framework. These MongoDB Atlas Vector Search to
features enabled us to provide the right drive efficiency through real-time,
context to the LLM and the appropriate data audio diagnostics.
to the AI model.

Ultimately, we obtain the desired output at

each stage, ranging from machine
prioritization, failure type identification, and
repair plan formulation, to instruction
generation.

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 Knowledge
Management

The Unique Challenges of workers, including both textual and unstructured

information. Creating semantic vectors from
Preserving Knowledge
these documents, manuals, and notes simplifies
Preserving and maintaining knowledge in
this process. The task of locating and transferring
manufacturing is just as much a challenge as
knowledge from data silos gets eliminated by
its accessibility. Due to aging population
indexing and generating vectors from a wide
worldwide, as experienced workers retire,
range of data sources, encompassing both
valuable tribal knowledge is lost. Transferring
structured documents and unstructured data
their expertise to the in-experienced workforce
such as handwritten notes. This enables users to
is difficult. Siloed data resulting from mergers
perform cross-system searches using natural
and acquisitions or legacy systems makes it
language queries facilitating seamless access to
even harder to consolidate knowledge for
information across different platforms.
decision making.

How Generative AI and Atlas Vector

Search Help
Manufacturers can capture and index the
valuable knowledge left by experienced

Preserving Expertise through MongoDB Atlas

Figure| 14:
US 866-237-8815 INTLPreserving knowledge
+1-650-440-4474 in MongoDB
| [email protected] Atlas Inc. All rights reserved.
| © 2023 MongoDB,
 Knowledge collection from shop With a knowledge management application on
the shopfloor all the input from the production
floor
equipment as well as from the workers can be
The time being wasted for consolidating data
collected. The application takes all the
from different systems to take decisions on a
structured and unstructured text input and
daily basis can be heavily reduced by using
categorizes it into one of many categories
gen AI. A shift leader, as an example, spends a
(defects, breakdowns, alarms, etc.).
lot of time to collects data from different
sources like MES, SCADA, or from handwritten
Having all that data collected, contextualized,
notes of the night shift workers to get an
and indexed allows a chatbot application to
overview about the condition of the equipment
get an immediate overview on the status of
after the last shift. Traditionally it takes a lot of
the shopfloor by prompting: Provide me a list
time to collect all that data from various
of machines with problems in the last shift,
sources and locations for getting a holistic
followed by prompting on how to solve that
overview to understand current safety,
problem. The RAG application can therefore
maintenance, inventory and quality needs.
use all the preserved information from the
experienced workers, stored as vectors in
A shifter leader may struggle to make timely
MongoDB Atlas.
decisions due to the fragmented nature of the
data, leading to delays in addressing
production issues and optimizing workflows.

Figure 15: Knowledge management architecture

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 Eni makes terabytes of subsurface With MongoDB Atlas, Eni users can quickly
unstructured data actionable with find data spanning multiple years and
MongoDB Atlas geographies to identify trends and analyze
Based in Italy, Eni is a leading integrated models. MongoDB Atlas Search also assists
energy company with more than 30,000 by filtering out irrelevant documents. The
employees across 69 countries. Its team also integrated AI and machine
operations vary from exploring and drilling learning models with the platform to make it
for natural gas and oil to cogenerating even easier to identify patterns.
electricity, renewables, biorefining, and
chemical production.
“MongoDB Atlas isnʼt just a
database, itʼs a complete set of
Eni partnered with MongoDB Consulting for
products and services. Itʼs cloud
training and to support the migration of
workloads into MongoDB Atlas. Eni wanted agnostic and combines rich
to move to a managed service with a functionality with the flexibility we
seamless user experience and easy-to-use
needed to make it our own.”
interface for developers.
Learn more
Sabato Severino
Senior AI Solution Architect for
Geoscience at Eni

Improving staff productivity at Enel The solution uses Amazon Titan, a family of
using Amazon Bedrock models exclusive to Amazon Bedrock.
Specifically, it uses the Amazon Titan Text
Enel is a leading integrated electric utility
Embeddings model to generate
with a presence across 32 countries and an
embeddings (vectors capturing semantics
82-GW generation capacity
of text) from Enel’s knowledge base, which
Enel identified the opportunity to use consists of a series of runbooks containing
generative AI to boost IT service desk incidents classes, preconditions, root
efficiency by extending automation to causes, resolutions steps, and operations
nontrivial tasks through basic information related to the applications.
troubleshooting, providing resolution steps Embeddings are computed and persisted in
and ticket routing without human a vector database instance using MongoDB
involvement. Atlas Vector Search, which supports
similarity search.
The solution is designed around a Learn more
retrieval-augmented generation
architecture using Amazon Bedrock.
US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
Other Notable Use Cases

AI plays a critical role in fulfilling the promise of Industry 4.0.

There are numerous other use cases of AI that can be enabled
by MongoDB Atlas.

Logistics Optimization Smart After Sales Support

AI can help optimize routes resulting in Manufacturers can utilize AI-driven chatbots
reduced delays and enhanced efficiency and predictive analytics to offer proactive
in day-to-day delivery operations. maintenance, troubleshooting, and
personalized assistance to customers.

Quality Control
Personalized Product
and Defect Detection
Recommendations
Computer or machine vision can be used to
AI can be used to analyze user behavior
identify irregularities in the products as they
and preferences to deliver personalized
are manufactured. This ensures that product
product recommendations via a mobile or
standards are met with precision.
a web app, enhancing customer satisfaction
and driving sales.
Production Optimization
By analyzing time series data from sensors
installed on production lines, waste can be
identified and reduced, thereby improving
throughput and efficiency.

Contact
Information
Dr. Humza Akhtar Raphael Schor
Manufacturing & Motion Manufacturing & Motion
Industry Solutions Principal Industry Solutions Principal
[email protected] [email protected]
 Atlas for Industries

Telecommunications
and Media
Faced with high
operational costs and
low margins, the
telecommunications and
media industries are
exploring new ways to
create value and enhance
revenue streams with AI.
AI investments in the telco & media industry

Generative AI is set to revolutionize the media These are some of the key findings from an
and telecommunications sectors, significantly industry survey of over 300 digital leaders
impacting content creation, distribution, across 50 countries:
service assurance, anti-fraud measures, and
overall industry dynamics. ● Referral Traffic: 63% are worried
about declining traffic from social
Content Creation and Distribution media, with significant drops reported
AI will dramatically enhance both content from platforms like Facebook and
creation and distribution. Search generative X/Twitter. Traffic to news sites from
experiences (SGE) and AI-driven chatbots are Facebook fell 48% in 2023, with traffic
expected to provide faster, more intuitive from X/Twitter declining by 27%.
access to information, shifting audience flows ● Platform Strategies: Publishers are
and financial dynamics within the industry. As prioritizing platforms like WhatsApp
referral traffic from traditional social media and Instagram while maintaining
platforms like Facebook and X (formerly interest in video networks such as
Twitter) declines, media companies face the TikTok and YouTube.
challenge of maintaining their audience and ● Content Focus: There is a strong
revenue streams. emphasis on increasing video,
newsletters, and podcasts to engage
In response to these challenges, media audiences, while concerns about news
organizations are focusing on building direct fatigue and selective avoidance persist.
relationships with their audiences and
investing in intellectual property protection. Revenue and AI Integration
This strategy aims to reduce dependence on Subscription and membership models are seen
major tech platforms but risks alienating as vital revenue streams, with publishers also
younger, tech-savvy audiences who are exploring lucrative AI platform licensing deals.
already comfortable with algorithmically AI is primarily viewed as a tool for backend
generated news. automation, improved recommendations, and
commercial uses.

Today’s leading companies in the

Opportunities and Strategies
telecommunications and media industry
Despite the challenges, the shift towards
are leveraging AI to enhance operations,
AI offers numerous opportunities. News
safeguard their businesses, deliver relevant
organizations are exploring innovative
content and services to customers, and
ways to integrate AI, emphasizing the
drive innovation.
importance of balancing technological
benefits with potential risks.
The telecommunications industry
operates in a landscape characterized
by tight profit margins, particularly in
commoditized communication and
connectivity services where
differentiation is minimal.

With offerings such as voice, data,

and internet access being largely
homogeneous, telecom companies need to
differentiate and diversify revenue streams
to create value and stand out in the market.

As digital natives disrupt traditional and Business Support Systems (BSS),

business models with agile and elevate customer service standards,
innovative approaches, established and enhance overall customer
companies are not only competing experiences to secure market share
among themselves but also with and gain a competitive edge. They’re
newcomers to deliver enhanced not alone—media publishers, too,
customer experiences and adapt to must streamline operations through
evolving consumer demands. automation while strengthening
reader relationships to foster a
To thrive in an environment where willingness to pay for personalized
advanced connectivity is increasingly and relevant content.
expected, telecom operators must
prioritize cost efficiency in their
Operations Support Systems (OSS)
 Service
Assurance

Telecommunications providers need to MongoDB Atlas’s JSON-based document

deliver network services at optimal quality model is the ideal data foundation to underpin
and performance levels to meet customer intelligent applications. It stores log data from
expectations and service level agreements. various systems without the need for
Key aspects of service assurance include time-intensive upfront data normalization
performance monitoring, quality of service efforts and with the flexibility to deal with a
(QoS) management, and predictive analytics wide variety of different data structures, as
to anticipate potential service degradation well as with their potential change over time.
or network failures. With the increasing
complexity of telecommunications networks By vectorizing the data with an appropriate ML
and the growing expectations of customers for model, it will be possible to reflect the healthy
high-quality, always-on services, a new bar has system state and to identify log information
been set for service assurance, requiring that shows abnormal system behavior. Atlas
companies to invest heavily in solutions that Vector Search allows for conducting the
can automate and optimize these processes required search in an effective way and as a
and maintain a competitive edge. fully included service
of the MongoDB cloud data platform.
Service assurance has been revolutionized by Finally, using LLM, information about the error,
AI through several key capabilities: ML can including the analysis of the root cause, can be
be the powerful foundation for predictive expressed in natural language, making the job
maintenance, analyzing patterns and of understanding and fixing the problem much
predicting network failures, allowing for easier for the staff who are in charge of
preemptive maintenance and significantly maintenance.
reducing downtime. AI techniques can also sift
through complex network systems to
accurately identify the root causes of issues,
improving the effectiveness of troubleshooting
efforts. AI can also have a huge impact on
network optimization, analyzing log data to
identify opportunities for improvement, raising
efficiency and thus reducing operational costs
and optimizing network performance in
real-time.
Fraud Detection
and Prevention

Telecom providers today are utilizing an The models can adapt over time, learning from
advanced array of techniques for detecting new data and emerging fraud tactics, thus
and preventing fraud, constantly adjusting to enabling real-time detection and the
the dynamic nature of telecom fraud. Routine automation of fraud prevention measures,
activities for detecting fraud consist of reducing manual checks, and speeding up
tracking unusual call trends and data usage, response times.
along with safeguarding against SIM swap
incidents, a method frequently used for To deal with fraud successfully, a multitude of
identity theft. To prevent fraud, strategies are data dimensions need to be considered, with
applied at various levels, starting with reaction time being a critical factor in
stringent verification for new customers, preventing the worst things from happening,
during SIM swaps, or for transactions with so the solution must also support fast,
elevated risk, taking into account the unique sub-second decisions. By vectorizing the data
risk profile of each customer. with an appropriate ML model, normal
(healthy) business can be defined, and in turn,
Machine learning offers telecommunications deviations from the norm identified, for
companies a powerful tool to enhance their instance, suspicious user activities. In addition
fraud detection and prevention capabilities by to Atlas Vector Search, the MongoDB Query
training ML models on historical data like Call API supports stream processing, simplifying
Detail Records (CDR). Moreover, these data ingestion from various sources.
algorithms can assess the individual risk
profile of each customer, tailoring detection One of the customers utilizing MongoDB for its
and prevention strategies to their specific anti-fraud strategy is AT&T. The company
patterns of use. selected MongoDB for its ability to ingest and
store rapidly changing data. Moreover,
MongoDB Atlas meets AT&T's key requirements
for performance, availability, and security. The
fraud prevention platform now operates over
50 different AI models, utilizes more than 1,200
features, stores 30TB of data, and processes
up to 20 million events daily. Events are
processed in under 200 milliseconds, enabling
real-time fraud detection and prevention. As a
result, fraud—particularly iPhone-related fraud,
which was previously a $1 billion problem—has
been reduced by over 80%.
Content Discovery
Content Discovery

Today’s media organizations are expected to Content Suggestions and Personalization

offer a degree of content personalization, from By utilizing user data, behavior analytics, and
streaming services to online publications and the multi-dimensional vectorization of media
more. Viewers want intelligently selected and items, the platform suggests content that
suggested content tailored to their interests. aligns with individual preferences and past
interactions. This not only enhances user
Leveraging AI can significantly enhance the engagement but also increases the likelihood
process of suggesting the next best article to of converting free users into paying
read or show to stream. The most powerful subscribers. Vectors are embedded directly in
implementations of content personalization MongoDB documents. This has several
track behavior of the user, like which content advantages. For instance, there are no
was searched for, how long was content complexities of a polyglot persistence
displayed before the next click happened, architecture, and no need to extract,
what categories, etc. Based on these transform, and load (ETL) data between
parameters, similar content can be presented, different database systems, which simplifies
or, as an alternative strategy, content from the data architecture and reduces overhead.
unseen areas of the portal presented to have Furthermore, MongoDB’s built-in scalability
the user discover new types of media and and resilience become particularly
check their appetite for consuming it. advantageous when handling vector search
operations. Organizations can scale their
To bring the right content to the right people at
operations vertically or horizontally, and they
the right time, an automated system needs to
can even choose to scale search nodes
maintain a multitude of information facets,
independently from operational database
which will lay the foundation for proper
nodes, thus adapting to the specific load
suggestions. With MongoDB and its document
scenario.
model, all required data points can be easily
and flexibly stored in a user’s profile, in Content Summarization and Reformatting
content, and media. In an age of information overload, the solution
provides concise summaries and adapts
Ultimately, by vectorizing the content, an even
content formats based on user preferences
more powerful system of content suggestions
and device specifications. This tailored
can be built with Atlas Vector Search, which
approach addresses the diverse consumption
allows for similarity search that goes well
habits of users across different platforms.
beyond comparing just keywords or a list of
attributes.
 Keyword Extraction Automatic Creation of Insights and
Essential information is drawn from content Summaries
through advanced keyword extraction, Our system can automatically generate
enabling users to grasp key news dimensions comprehensive insights and summaries from
quickly and enhancing the searchability of multiple articles. This feature is particularly
content within the platform. Keywords are valuable for users interested in deep dives into
fundamental to how content is indexed and specific topics or events, providing them with a
found in search engines, and they significantly rich, contextual experience. This capability
influence the SEO (search engine optimization) leverages the power of one or more large
performance of digital content. With the help language models (LLMs) to generate natural
of the underlying LLM, the solution extracts language output, enhancing the richness and
keywords automatically and with high accessibility of information derived from
sophistication. across multiple source articles.

Solution demo
Discover how to leverage MongoDB
and Vector Search effectively to
create more engaging, personalized
user experiences.

Figure 16: Reference architecture for to creating more engaging, personalized user experiences with
MongoDB and Vector Search
Search Generative
Content Discovery
Experiences (SGE)

Search generative experiences (SGE) represent the most relevant articles for a topic of
a transformative shift in how media and interest, but also combining them into a
content industries interact with information concise summary, elevating the value for the
retrieval and user engagement. SGE utilizes
customer.
artificial intelligence to generate search results
that are not just links to existing content, but
Smart Conversational Interfaces
comprehensive, synthesized responses tailored
to the user's query. By providing immediate, Content retrieval products that are accessible
concise answers, SGE helps media platforms from smartwatches, AR and VR glasses, lapel
keep users engaged longer, reducing bounce pins, and other wearable devices, or from the
rates and increasing the likelihood of deeper car, allow for the consumption of data in more
interaction with the content. day-to-day situations.

With MongoDB and Vector Search being

Gamified Learning Experiences
placed at the center of an SGE solution (as
shown below in the reference architecture), Develop educational content with interactive
multiple new revenue streams can be quizzes and activities that adapt to the user's
established. knowledge level and learning pace using Atlas
Vector Search and SGE to provide personalized
Intelligent Research Tools feedback.
Provide easy access for researchers to the
publisher’s news archive, returning not just

Figure 17: Reference architecture for creating Search Generative Experiences

 Other Notable Use Cases

Differential Pricing Backend Automation

Gather insights into what customers are Extract relevant keywords from news articles
willing to spend on content or a service by using one or several LLMs, thus supporting
conducting A/B tests and analyzing the data search engine optimization (SEO), as well as
with an ML algorithm. content indexing and categorization.

This method facilitates the adoption of Assist in copyediting, by identifying and

dynamic pricing models instead of sticking to correcting grammatical errors, spelling
a standard price list, thereby increasing mistakes, and punctuation issues. Suggest
overall revenue, and enlarging the paying improvements in sentence structure, style,
customer base. and tone, ensuring high-quality content
production.

Vodafone transformed its software development with a

"telco-as-a-service" (TaaS) model and MongoDB Atlas, enabling
Vodafone’s New Developer Speed faster, more secure app development. This shift to serverless

and Dexterity—Powered by computing cut costs, improved customer experience, and

boosted developer productivity, allowing teams to quickly
MongoDB
deploy code and innovate.

Contact
Information
Benjamin Lorenz
Telco & Media Industry
Solutions Principal
[email protected]
 Atlas for Industries

Retail

AI is transforming retailers’
ability to maximize their
competitive advantage
through better understanding
of their customers and
improving their operating
margins through intelligent
decision making.
Artificial Intelligence (AI) is revolutionizing the
retail industry across the globe, driving
innovation and enhancing efficiency. AI itself is
evolving from traditional AI to generative AI.

The shift from AI to generative AI in retail

reflects advancements in technology that
enable more sophisticated and creative
applications, improving customer experiences
and operational efficiencies.

Traditional AI (machine learning models and create personalized shopping

arithmetic algorithms) have been used experiences by dynamically generating
extensively in retail for a variety of functions: web and mobile interface elements
based on user behavior.
● Personalization: AI-driven ● Conversational Chat: AI-generated
recommendation engines analyze virtual shopping assistants can provide
customer data to provide personalized more natural and engaging interactions
product suggestions. with customers, improving the overall
● Demand Forecasting: Predictive shopping experience.
analytics help retailers manage inventory
by forecasting demand and optimizing Major consulting firms have extensively
stock levels. documented these advancements, for
example, in a 2023 survey by McKinsey
Generative AI represents a leap forward by about one third of all respondents said their
not only analyzing data but also creating organizations were already regularly using
new content and solutions: generative AI in at least one function
(McKinsey).
● Content Creation: Generative AI can
produce personalized marketing content, MongoDB sees AI as having a
such as product descriptions, transformative impact on global retail by
advertisements, and social media posts, driving innovation and enhancing customer
tailored to specific audiences. experiences. Leveraging MongoDB Atlas and
● Hyper Personalization: Beyond its integration with different platforms,
recommendations, generative AI can retailers can manage massive datasets
 required for generative AI applications
effectively. This enables advanced data
With the use of AI, retailers can
ingestion, seamless AI model training, and
leverage the advantages of
efficient data retrieval through features like
augmented and semantic search,
vector search. These capabilities allow retailers
generate marketing materials
to automate tasks, personalize customer
based on the market conditions,
interactions, and innovate with new content
formats, ultimately leading to faster
get the most of predictive
time-to-market and cost-effectiveness. analytics to forecast demand,
McKinsey & Company estimates generative AI and use conversational chatbots
could contribute roughly $310 billion in to enhance customer
additional value for the retail industry by experiences.
boosting performance in functions such as
marketing and customer interactions
(McKinsey). The most creative AI use cases for retailers is
to understand customer needs and choices
Generative AI has significantly impacted the that change continually with season, trends,
retail industry, driving notable growth and and socio economic shifts. By analyzing
transformation. According to NVIDIA's "State customer data and behavior, generative AI
of AI in Retail and CPG" survey, 69% of can also create personalized product
retailers reported an increase in annual recommendations, customized marketing
revenue due to AI adoption, and 72% materials, and unique shopping experiences
experienced a decrease in operating costs. that are tailored to individual preferences.
Generative AI is particularly influential in
enhancing customer experiences, with 86% of AI plays a critical role in decision making at
retailers recognizing its potential in this area retailer enterprises—product decisions such as
(Nvidia). design, pricing, demand forecasting,
and distribution strategies require complex
understanding of a vast amount of information
from across the organization.

Retail leaders who invest more in To ensure that the right products in the right
technology to improve customer quantities are in the right place at the right
experience are 17% more likely to time, back office teams leveraged machine
outperform other retailers in learning arithmetic algorithms for years.
organic sales growth and get
about 37% more sales from As technology has advanced and the barrier
digital channels (Bain and for entry is lowered for adopting AI, retailers

Company). are moving towards data-driven decision

making where AI is leveraged in real time.
Generative AI is used to consolidate
information and provide dramatic insights that
could be immediately utilized across the
enterprise.
US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
AI-Augmented Search
and Vector Search

Retail is a customer centric business. analytics from AI algorithms. Also, it's now
Customers have more choice than ever in possible to incorporate factors such as the
where they purchase a product. To retain and current shopping cart or basket and customer
grow their customer base, retailers need to clickstream and trending purchases across
keep innovating in order to offer each shoppers.
customer a differentiated buying experience.
To do this, it is necessary to use a large The first step in truly understanding the
amount of data from the customers such as customer is to build a customer operational
buying patterns, interests, and interactions data store that combines data from disparate
and to be able to quickly make complex systems and silos in the organization: support,
decisions on that data. e-commerce transactions, in-store interactions,
wish lists, reviews, and more. MongoDB’s
One of the key customer interactions in an flexible document model enables bringing
e-commerce experience is search. Through the data of different types and formats in one
implementation of full-text search engines, document to get a clear view of the customer
customers can more easily find items that in one place. As the retailer captures more
match their search, and retailers are given the data points about the customer, they can
opportunity to rank those results in a way that easily add fields without the need for
will give the customer the best option. downtime due to schema changes.
Traditionally, decisions on how to rank search
results in a personalized way were made by Then comes the ability to run analytics in
segmentation of customers through data real time rather than retroactively in another
acquisition from various operational systems, separate system. MongoDB’s architecture
moving it all into a data warehouse, and allows for workload isolation, meaning
subsequently running classical AI with various operational workloads (the customer's actions
machine learning algorithms on such data. on the ecommerce site) and the analytical or
Typically, this would run in a batch mode AI workload (calculating what the next best
(every 24, 48, or even 72 hours or a few days), offer should be) can be run simultaneously
and the next time a customer logs in, they will without interrupting the other. Retailers can
have a personalized experience. It does not, build dynamic ranking by using the MongoDB
however, capture the customer's true desire in aggregation framework for advanced
real time. analytical queries or triggering an AI model in
real time to give an answer that can be
Modern retailers augment search ranking with embedded into the search ranking.
data from real-time responses and/or

41
The benefit of an all-in-one platform is huge
here, as instead of having to update your
Vector search technology in retail
search indexing to incorporate your AI
provides notable economic
augmentation, MongoDB has Lucene Search
benefits. As highlighted by
built in. This whole flow can be completed in
Deloitte: Sales Uplift and
one data platform automatically—as your data
Customer Engagement, Deloitte
is being augmented with AI results, the search
indexing will sync to match.
reported that retailers
implementing personalized
MongoDB Atlas Vector Search brings the next search have seen a sales uplift of
generation of search capability. By using LLMs about 40% (Deloitte).
to create vector embeddings for each product
and then turning on a vector index, retailers
are able to offer semantic search to their
customers. AI will calculate the complex
similarities between items in vector space and
give the customer a unique set of results READ MORE
matched to their true desire. AI-Enhanced Search in
E-commerce With MongoDB

Figure 18: Architecture of an AI-enhanced search engine explaining the different MongoDB Atlas
components and Data Intelligent Platforms and workflows used for data cleaning and preparation,
product scoring, dynamic pricing, and vector search
Delivery Hero Helps Customers The solution was MongoDB Atlas, a
Navigate more than 100 Million scalable, high-performing developer data
Products with MongoDB Atlas Search platform that integrates easily with the AI
ecosystem. With it the new Item
Replacement Tool is being piloted first in the
Delivery Hero, a food delivery service
Middle East. By providing personalized
based out of Germany, has built a new Item
recommendations against live inventory,
Replacement Tool providing
Delivery Hero expects to see an increase in
hyper-personalized product
its monthly gross merchandise value.
recommendations in real time using
state-of-the-art AI models and MongoDB
Atlas Vector Search. “We can implement things that
would take six months on other
The challenge was that around 10% of the
platforms in just a few days.”
inventory is perishable produce that can
quickly go out of stock. Without being able Learn more
to recommend a suitable alternative to the Andrii Hrachov
customer, the company risks revenue loss Principal Software Engineer at Delivery
and customer churn. Hero

Personalized
Advertising, and marketing material are vital
to capturing a customer’s interest and driving

Marketing towards a purchase. With the advent of social

media there are now many more ways to reach
& Content the customer than before: Instagram,

Generation Facebook, email outreach, newsletters, and

promotional banners on sites. This creates a
lucrative opportunity for retailers but also a
challenge when it comes to a huge amount of
content generation.
Customer buying patterns, constantly updating
product catalog, and inventory availability are
critical components of retail operations. Along
with this there is also the task of ensuring that
the product literature is in the right tone of
voice to reflect the brand in multiple
languages. The product images need to be
relevant to the audience in the locale.
Traditionally, this required a huge amount of
labor in copywriting and editing, photography
of different models, and generation of visuals
and graphics.

The retailer must also understand in real time

what the impact of campaigns is so they can
quickly redirect their marketing spend and
strategy to reflect what is working. In an
industry where marketing and branding are
Gen AI has also rapidly increased retailers'
critical business activities, retailers need as
ability to personalize the interactions with their
much insight on their customers as possible so
customers. Retrieval-augmented generation
they can reach them with the right message at
using Large Language Models (LLMs) is
the right time.
capable of creating individualized marketing
material, newsletters, social posts, and email
outreach that is unique to each customer in
seconds. Visuals, graphics, and even
Companies will take advantage of the photo-realistic images can be generated using
sharp rise in consumer touchpoints to AI to leverage the vast amount of data the
personalize and reach the growing retailer already has. This reduces manual work
population of consumers who use digital and accelerates time to market.
channels to discover, consider, and
AI can also be used to understand quickly and
purchase products. Sixty-five percent of
easily the effectiveness of campaigns, giving
consumers research products online,
insights to drive intelligent, strategic decisions.
and 30% buy online. These numbers
have doubled over the past three to four The key to creating content that is
years. This creates an enormous need personalized to the customer and the brand
for brands to target online consumers is leveraging the vast amount of data that
with personalized content—an retailers have in-house to provide an LLM
opportunity enabled by generative AI’s with context.
lower content creation costs (Bain and
Company).
In MongoDB, the Apache Spark Connector With the MongoDB platform approach, as new
allows for model training of LLMs items are added to the product catalog, or
so that prompts such as “create a personalized new images and visuals, change streams can
newsletter for each customer suggesting be used to trigger the vectorization of new
an item based on what is on offer and their data so that the process becomes seamless.
previous purchases” can use data, images, and Training the model with your internal data
tonal or language references to create provides an invaluable resource to retailers in
outreach. reaching their audience easily

Figure 19: AI-powered personalization architecture

The above shows a reference architecture By doing so, media organizations can suggest
highlighting where MongoDB can be leveraged content that aligns more closely with individual
to achieve AI-powered personalization. By preferences and past interactions. This not only
leveraging user data and the enhances user engagement but also increases
multi-dimensional vectorization of media the likelihood of converting free users into
content, MongoDB Atlas can be applied to paying subscribers.
multiple AI use cases. This allows for the
utilization of media channels to more
effectively improve end-user experiences.
Figure 20: Example of the data flow for an AI-generated personalized newsletter. The prompt is
entered by a user on the left hand side and context is added via the vectorized data in
MongoDB—wishlist, loyalty data, purchase history, and customer behavior. Using RAG, the LLM
can produce a personalized newsletter per customer in seconds, allowing the retailer to create
vast amounts of customized content

Accurate demand planning using AI in retail

Demand optimizes inventory levels, reducing costs and

stockouts while enhancing customer
Forecasting satisfaction through better availability of

& Predictive
products. It also enables data-driven
decisions, leading to improved sales forecasts

Analytics and efficient supply chain management.

Retailers either develop homegrown
applications for demand prediction using
traditional machine learning models or buy
specialized products designed to provide
these insights across the segments for
demand prediction and forecasting. The
homegrown systems require significant
infrastructure for data and machine learning
implementation and dedicated technical
expertise to develop, manage, and maintain
them. More often than not, these systems
require constant care to ensure optimal
performance and provide value to the
businesses.
Subsequently, feature engineering is there to Such efforts lead to the following business
extract seasonality, promotions, impact, and benefits:
general economic indicators. A retrieval
augmented generation model can be ● Precision in demand forecasting
incorporated to improve demand forecasting ● Optimized product / supply planning
predictions and reduce the possibility of ● Accuracy in inventory management
hallucinations. The same datasets could be ● Enhanced customer satisfaction
utilized from historical data to train and
fine-tune the model for improved accuracy.

Traditional AI is used in demand forecasting

and predictive analytics in retail by Well implemented demand
integrating data from diverse sources like
forecasting can lead to a 3-7%
sales transactions, social media, and
weather patterns, leading to highly accurate
increase in group operating
and timely forecasts. Machine learning profit, 30% reduced time to
algorithms continuously learn and adapt market 15-20% increase in store
from new data, improving predictive ordering daily, increasing product
accuracy, while automation reduces the
freshness (Bain and Company).
time and resources needed for these tasks,
allowing for efficient scaling of forecasting
efforts.

Generative AI is transforming demand

forecasting and predictive analytics to find
patterns from existing datasets, enhancing
the accuracy and depth of predictions. By
creating synthetic data, generative AI
models can fill in gaps in historical data,
simulate various market scenarios, and
predict future trends more effectively. This
leads to more precise demand forecasts,
allowing retailers to optimize inventory
levels, reduce stockouts, and avoid
overstock situations, thus improving
operational efficiency and customer
satisfaction.
Figure 21: An illustration of a price-change scenario where fuel costs have risen, which leads to a
rise in shipping costs and, in turn, pricing

● [1] This produces events about the cost increase and places them" with "The price
increases are placed in the message stream where the event queue makes them available.
All microservices are listening for such messages.
● [2-3-4] The pricing microservice consumes the event, analyzes it against existing data,
and further conveys the new pricing into the message stream.
● [5 - 6] The database pushes those messages to the event queue, which makes them
available to all consumers listening for messages. Microservices directly impacted by
pricing changes—such as those that manage inventory, marketing, promotions, coupons,
point of sale (POS), and the e-commerce provider’s order management system
(OMS)—consume the price change events and update their individual databases
accordingly.
● [7] The centralized database aggregates and persists events, enriches event streams with
data from other sources, including historical data, and provides a central repository for
multiple event streams.
 Conversational Conversational chatbots powered by
generative AI are revolutionizing the retail
Chatbots industry by enhancing customer service.
These chatbots can handle a wide range of
customer inquiries, from product
recommendations to order tracking,
providing instant and accurate responses.
This reduces wait times and improves the
overall customer experience, leading to
higher satisfaction and increased loyalty.
Additionally, chatbots can operate on real-
time data 24/7, ensuring customers receive
support at any time, which is especially
beneficial for global retailers.

Beyond customer service, AI chatbots are also Operational efficiency is another area where
transforming marketing and sales strategies in AI chatbots are making a significant impact.
retail. They can analyze customer data to By automating routine tasks such as
personalize shopping experiences, offering answering FAQs, managing inventory inquiries,
tailored recommendations and promotions and processing returns, chatbots free up
based on individual preferences and behavior. employees to focus on more complex and
This personalization helps retailers boost value-added activities. This not only reduces
conversion rates and increase sales. Moreover, operational costs but also improves accuracy
chatbots can engage customers through and consistency in service delivery.
various digital channels, including social Furthermore, the data collected by chatbots
media, websites, and messaging apps, can provide valuable insights into customer
broadening the reach and effectiveness of preferences and behavior, helping retailers
marketing campaigns. refine their strategies and improve their
offerings.

Across both savvy and non-savvy digital users, 50%-60% have

shown high preference to move to conversational journeys for
day-to-day use cases across verticals (Bain and Company).
Following is a chatbot RAG architecture To retrieve relevant information based on user
example. This chatbot is built using the queries, MongoDB Atlas Vector Search is
retrieval-augmented generation (RAG) utilized. In this example OpenAI is being used
architecture. RAG augments the knowledge of in tandem with Vector Search to generate
large language models (LLMs) by retrieving answers to the customer questions. Using data
relevant information for users' queries and from private data sources and enhanced by
using that information in the LLM-generated the LLMs the data is augmented, given
response. MongoDB's public documentation is context, and then returned to the user. The
used as the information source for chatbot Azure OpenAI embeddings API are used to
generated answers. convert MongoDB documentation and user
queries into vector embeddings to help find
the most relevant content for queries using
Atlas Vector Search.

Figure 22: Example of the data flow for a chatbot RAG architecture
 L’Oréal Improves App Performance The result was reducing latency from
and Velocity with MongoDB Atlas seconds to just 10 milliseconds.

The challenge for L’Oréal was to complete “MongoDB Atlas doesnʼt just solve
complex calculations on vast volumes of our performance issues It makes
data—without causing latency. The solution life easier. We have a hyper agile
was simplifying management and
maintenance while boosting performance
DevOps model.”
with MongoDB Atlas. Learn more
Moutia Khatiri
CTO, Tech Accelerator, L’Oréal

Artificial intelligence is revolutionizing the way benefits of enhanced and semantic search
retailers enhance their competitive edge by capabilities, create targeted marketing content
providing deeper insights into customer based on current market trends, effectively
behavior and optimizing profit margins utilize predictive analytics for demand
through smart decision-making processes. By forecasting, employ conversational chatbots,
incorporating both traditional and generative and significantly elevate the overall customer
AI, retailers can harness the experience.

Contact
Information
Genevieve Broadhead Prashant Juttukonda
MongoDB Global Lead, Retail Industry
Retail Solutions Solutions Principal
[email protected] [email protected]

Rodrigo Leal
Retail Industry
Solutions Principal
[email protected]
 Atlas for Industries

Financial
Services
Analyst firms, Celent and
McKinsey, both identify
financial services as one
of the industries most
likely to be transformed
AI. While the initial
benefits are likely to be
process-related as firms
slowly ramp up adoption,
it won't be long before
customer interactions are
also completely
re-invented by AI-driven
experiences.
 AI investments in the financial services industry

Already an integral part of the industry in The conversation around generative AI

areas such as risk, fraud, and compliance for became more nuanced through the latter part
example, the use of AI has been commonplace of 2023. This is understandable given the
for years and continues to deepen, while there complexities of applying large language
is a long (and growing) list of other use cases models (LLMs) to sensitive customer data, as
for which banks are also making use of AI. well as broader regulatory concerns over the
explainability of LLM outputs. That said, there
AI technologies are used today to address a are many areas in which gen AI is already
wide range of different workflows and being used to support advisors and
customer-facing services from process relationship managers, with further innovation
automation and optimization in the middle and expected. According to Celent’s Harnessing the
back office, to areas such as real-time risk and Benefits of AI in Payments report, 58% of
service personalization in the front office. banks are evaluating or testing generative AI
However, it is to be expected that the benefits in some capacity (as shown below) while a
will be first in internal process optimizations further 23% have projects using this in their
and will only slowly start to be seen in direct roadmap.
customer engagement and interaction.

Interest in leveraging AI
for banking use cases is
expected to increase in
the next few years.

Figure 23: Bank tech exploration vs. expected impact

Addressing the challenges of AI in finance

While the industry has always had to deal with A strategic and comprehensive approach
persistent issues like risk management and encompassing aspects of technology, data,
governance, adopting generative AI and ethics, and organizational readiness is critical
machine learning introduces new challenges to overcoming these challenges:
that AI specialists have always dealt with, like
inherent biases and ethical concerns. One ● Hallucination mitigation: One promising
challenge that stands out for generative AI is strategy is using retrieval augmented
hallucination—the generation of content that is generation (RAG) to mitigate hallucination
not accurate, factual, or reflective of the real in gen AI models–incorporating
world. information retrieval mechanisms to
enhance the generation process to ensure
Generative AI models, especially in natural content is grounded in real-world
language processing, might generate text that knowledge. Atlas Vector Search is a
is coherent and contextually appropriate but popular mechanism to support RAG,
lacks factual accuracy. This poses challenges, which uses vector embeddings to retrieve
for example: relevant documents based on the input
query.
● Misleading financial planning advice: ● Data quality and availability: Take a
In financial advisory services, hallucinated step back before adopting AI to ensure
information may result in misleading the quality, relevance, and accuracy of
advice, leading to unexpected risks or data being used for AI training and
missed opportunities. decision-making can be accessed in real
● Incorrect risk assessments for lending: time.
Inaccurate risk profiles may lead to poor ● AI education: The key is to invest in
risk assessments for loan applicants, training programs to address skill gaps,
which can cause financial institutions to create a culture of learning and
approve a loan with a higher risk of development, and promote awareness
default than the firm would normally about vulnerabilities.
accept. ● Develop new governance, frameworks,
● Sensitive information in generated and controls: Before going live, create
text: When generating text, models may safe and secure environments for testing.
inadvertently include sensitive information ● Implement monitoring systems:
from the training data. Adversaries can Measure and understand financial
craft input prompts to coax the model impacts, change impacts, scale, and
into generating outputs that expose complexity associated with AI.
confidential details present in the training ● Security and privacy: Implement secure
corpus. data access privileges and authentication
measures to safeguard AI models and the
data they rely on.
 Relationship Management
Support with Chat Bots

One key service that relationship managers ● When are the next coupon dates for
provide to their private banking customers bonds in the portfolio?
is aggregating and condensing information. ● What has been the cost of
transactions for a given portfolio?
Because banks typically operate on ● What would be a summary of our
fragmented infrastructure, this can latest research?
require a lot of detailed knowledge ● Please generate a summary of my
about this infrastructure and how conversation with the client.
to source information such as:

Traditionally, relationship managers (RMs) There are different approaches of building a

faced challenges accessing and analyzing RAG-based chatbot application.
client portfolio data. Tasks like finding
coupon dates were time-consuming and A simple approach would be like the one
required manual effort. To streamline this explained in our article Taking RAG to
process, RMs often requested new reports, Production with the MongoDB
but finding the right information could be Documentation AI Chatbot with an
difficult. architecture diagram as follows.

Generative AI offers a solution by

automating data retrieval and analysis.
Through RAG, AI can provide answers and
explanations, and even execute queries
based on diverse data sources, including
portfolio information, user manuals, and
customer data. RMs can interact with the AI
to refine results and ensure accuracy.

MongoDB's platform is well suited for this

Figure 24: User query processing flow
application. It can handle various data types,
including structured and unstructured The more advanced approach would be to
information, and it provides efficient data leverage a composable RAG framework that
search and processing tools. This enables the is developed as part of the MongoDB AI
development of AI-powered applications that Applications Program (MAAP).
deliver timely insights to RMs.
Figure 25: Query processing pipeline

Data loading from raw data sources mentioned text chunks into vector
Begin by extracting data from databases or representations. The chunks can then
files. Examples of data can include customer be indexed using MongoDB vector
profiles, transaction histories, and product indexing that will then enable an
information. The source of the data may be efficient retrieval of vectorized data
located in different databases or files and can using Atlas Vector Search.
be consolidated into MongoDB and serve as
an operational data store to provide a Pre-query Retrieval
customer 360-degree view. Advanced RAG methods explore various
transformations of user queries to enhance
Data indexing: chunking and vectorization retrieval accuracy before it is passed to the
● Data chunking involves breaking LLM for response generation. As an example, If
down the documents (such as the RM asks for all recent fund transfers for a
customer service records, emails, and given customer, the gen AI application can
transaction logs) into smaller, more pre-process the query by first checking if the
manageable chunks (eg. individual RM has access privileges to the customer data.
interactions or sentences) and then It can then refine the query to add a specific
creating an index that allows for quick time period, and add the relevant documents
searching and retrieval of relevant retrieved from the vector search. This will
information. augment the search, giving it context to
● Data vectorization is implemented by optimize relevancy of the results.
using the embedding model (eg.
Cohere, Anthropic, etc) to convert the
above
Post retrieval also be trained to specialize in a specific
With the transformed query executed, the domain for accuracy and efficiency. For
relevant documents will be retrieved (such as example in the context of financial advisory
with regular, text, vector or even hybrid support, an agent could specialized in
searches). The retrieved documents (such as Natural Language Understanding of
the customer details and transactions) can be financial market information, another AI
further filtered and reranked to provide an Agent can specialist in Risk Assessment for
even more relevant set of documents to be evaluating client and product risk profiles
used as a data augmentation when prompting while yet another can specialized in
the LLM later. Compliance for regulatory adherence to
ensure the investment decisions especially
LLM for corporate clients are also compliant.
The LLM engine can be a single or a group of Together they can be “composed” using
LLM engines deployed on an AI platform (eg. orchestration tools (eg. Langchain and
Fireworks.AI) to work together leveraging LLamaIndex) to support the tasks required
different specialization to provide a relevant, for the generative AI interactions.
accurate and combined response to the users.
Let us generalize and refer to these engines as
AI agents so to not limit our imagination to
just Language Models (be it large or small).
These AI Agents not only can perform
orchestration tasks but can agent can
Risk Management
and Regulatory Compliance

Risk & Fraud

Prevention

Banks are tasked not only with particular, has shown promise in trade
safeguarding customer assets but also with surveillance, offering a more dynamic and
detecting fraud, verifying customer comprehensive approach to fraud
identities (KYC), supporting sanctions prevention.
regimes, and preventing various illegal
activities (AML). The challenge is magnified Modernizing fraud systems improves
by the sheer volume and complexity of detection accuracy, increases flexibility and
regulations, making the integration of new scalability, enhances security, and reduces
rules into bank infrastructure costly, the cost of operation. MongoDB's developer
time-consuming, and often inadequate. For data platform is a great fit for modern
instance, when the EU's Fifth Anti-Money fraud prevention systems due to:
Laundering Directive was implemented,
Data visibility: MongoDB’s document
it broadened regulations to cover virtual
model incorporates any kind of data no
currencies and prepaid cards. Banks had
matter how often it changes, allowing you
to swiftly update their onboarding
to create a holistic picture of customers to
processes and software, train staff, and
better predict transaction anomalies in real
possibly update their customer interfaces to
time. As well as process large amounts of
comply with these new requirements.
data and analyze in real-time to identify
AI offers a transformative approach to suspicious activity. Change Streams in
fraud detection and risk management by MongoDB enable the monitoring of
automating the interpretation of database changes in real time meaning
regulations, supporting data cleansing, and immediate detection of suspicious activity
enhancing the efficacy of surveillance or transactions. By receiving instant
systems. Unlike static, rules-based notifications, fraud prevention teams can
frameworks that may miss or misidentify quickly respond to potential threats.
fraud due to narrow scope or limited data,
Real-time data processing: By leveraging
AI can adaptively learn and analyze vast
MongoDB as the operational data store
datasets to identify suspicious activities
(ODS) you can handle large amounts of
more accurately. Machine learning, in
data to effortlessly capture, store, and
 process high-volume transactional data in Besides the use of AI/ML techniques, there
real-time. MongoDB’s aggregation the option of leveraging vector data store
framework allows for complex data capabilities such as MongoDB Atlas Vector
processing and analytics directly within the Search to perform fraud and money
database, crucial for implementing laundering detection. This can be also be
sophisticated, rules-based fraud detection used in conjunction with the AI/ML
algorithms. approach to create a hybrid
implementation to further enhance the
Integrating with modern AI/ML accuracy of detection and minimize false
technologies: MongoDB’s flexible data positives. Learn More
model makes it easier to integrate with
AI/ML platforms to develop models for Below is an example architecture of an
real-time, predictive fraud prevention that ML-based fraud solution leveraging
are trained on the most accurate and MongoDB and Databricks. Key features
relevant data available. include data completeness through
integration with external sources, real-time
Regulatory compliance: Compliance with processing, AI/ML modeling to identify
regulations like PCI, GDPR, CCPA, and PSD2 potential fraud patterns, real-time
is crucial in the finance sector. MongoDB's monitoring for instant analysis, and model
security features ensure data handling observability for full data visibility into
meets these regulatory standards. fraud behaviors. Learn more
Enhanced authentication and role-based
authorization controls further fortify the
system.

Figure 26: Example fraud detection solution leveraging MongoDB

 Financial Document
Search and Summarization

Financial institutions, encompassing both retail MongoDB can store vast amounts of both live
banks and capital market firms, handle a and historical data, regardless of its format,
broad spectrum of documents critical to their which is typically needed for AI applications.
operations. Retail banks focus on contracts, It offers vector search capabilities essential for
policies, credit memos, underwriting retrieval-augmented generation (RAG).
documents, and regulatory filings, which are MongoDB supports transactions, ensuring
pivotal for daily banking services. On the other data accuracy and consistency for AI model
hand, capital market firms delve into company retraining with live data. It facilitates data
filings, transcripts, reports, and intricate data access for both deterministic algorithms and
sets to grasp global market dynamics and risk AI-driven rules through a single interface.
assessments. MongoDB boasts a strong partnership
ecosystem, including companies like Radiant
These documents often arrive in unstructured AI and Mistral LLM, to speed up solution
formats, presenting challenges in efficiently development.
locating and synthesizing the necessary
information. While retail banks aim to
streamline customer and internal operations,
Discover more about MongoDB’s
capital market firms prioritize the rapid and
full full-text search and vector
effective analysis of diverse data to inform
search capabilities.
their investment strategies. Both retail banks
and capital market firms allocate considerable
time to searching for and condensing
information from documents internally, To achieve search and summarization using
resulting in reduced direct engagement with RAG, a conceptual architecture leveraging
their clients. MongoDB and generative AI can be used.

Generative AI can streamline the process This architecture aims to centralize document
of finding and integrating information from storage, enhance search capabilities, and
documents by using NLP and machine automate the extraction and synthesis of
learning to understand and summarize information. The following diagram illustrates
content. This reduces the need for manual this conceptual architecture.
searches, allowing bank staff to access
relevant information more quickly.

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
Figure 27: Atlas hybrid search workflow: query embedding, document retrieval, summarization, user
presentation

ESG Analysis

The profound impact of environmental, banks to also improve their green lending
social, and governance (ESG) principles is practices. This shift is strategic for financial
evident, driven by regulatory changes, institutions, attracting clients, managing
especially in Europe, compelling financial risks, and creating long-term value.
institutions to integrate ESG into investment
and lending decisions. Regulations such as However, financial institutions face many
the EU Sustainable Finance Disclosure challenges in managing different aspects
Regulation (SFDR) and the EU Taxonomy of improving their ESG analysis. The key
Regulation are examples of such directives challenges include defining and aligning
that require financial institutions to consider standards and processes, and managing
environmental sustainability in their the flood of rapidly changing and varied
operations and investment products. data to be included for ESG analysis
Investors' demand for sustainable options purposes.
has surged, leading to increased
ESG-focused funds. The regulatory and
commercial requirements in turn, drive
AI can help to address these key challenges
in not only an automatic but also adaptive
manner via techniques like machine learning.
Financial institutions and ESG solution
providers have already leveraged AI to
extract insights from corporate reports,
social media, and environmental data,
improving the accuracy and depth of ESG
analysis. As the market demands a more
sustainable and equitable society, predictive
AI combined with generative AI can also help
to reduce bias in lending to create a fairer and
more inclusive financing while improving the
predictive powers. The power of AI can help
facilitate the development of sophisticated
sustainability models and strategies, marking
a leap forward in integrating ESG into broader
Incorporating LLMs enhances MongoDB's
financial and corporate practices.
capacity to process ESG textual content,
automating sentiment extraction,
MongoDB's dynamic architecture
summarization, and trend identification.
revolutionizes ESG data management,
Combining LLMs with vector data
handling semi-structured and unstructured
management capabilities, generative AI
data. Its flexible schema nature allows the
applications can be created to interpret the
adaptation of data models as ESG strategies
complex and evolving sustainability taxonomy
evolve. Advanced text search capabilities
and guide the investment and financing
efficiently analyze vast semi-structured data
processes in a compliant manner. This
for informed ESG reporting. Support for vector
AI-driven approach, supported by MongoDB's
search enriches ESG analysis with multimedia
robust data management, offers a
content insights.
sophisticated means of analyzing extensive
narrative data in ESG reporting.

Furthermore, MongoDB supports geospatial

and network graph analytics, providing a
powerful combination of analytics to identify
the physical risks associated with climate
change (e.g., floods, wildfires) to assets
financed by banks or investment firms and for
assessing supply chain impacts of the climate
risks. The risk analytics can then enable
targeted strategies for risk mitigation and
supply chain resilience.
MongoDB's value extends beyond ESG data
management, accelerating productivity for
developers and data science teams. Its
intuitive data model, analytical tools, and AI
integrations streamline the development and
deployment of data-driven applications,
making MongoDB pivotal for organizations
advancing their ESG agendas efficiently.

Below is a diagram of an enterprise ESG

solution architecture with the boxes labeled
with a leaf where MongoDB can be deployed
to support the ESG data analytics services.

Learn how MongoDB makes it

incredibly easy to collect, analyze,
and visualize ESG data in real time.

Figure 28: Blueprint for Enterprise ESG Solution Architecture Using MongoDB
 Transforming Credit
Scoring with AI

The convergence of alternative data, artificial technological capabilities but fundamentally

intelligence, and generative AI is reshaping the redefining how credit is evaluated, fostering a
foundations of credit scoring, marking a new era of financial inclusivity, efficiency, and
pivotal moment in the financial industry. The fairness.
challenges of traditional models are being
overcome by adopting alternative credit The use of artificial intelligence, in particular
scoring methods, offering a more inclusive and generative artificial intelligence, as an
nuanced assessment. Generative AI, while alternative method to credit scoring has
introducing the potential challenge of emerged as a transformative force to address
hallucination, represents the forefront of the challenges of traditional credit scoring
innovation, not only revolutionizing methods for several reasons.

Alternative data analysis: Unlike Predictive analysis: AI algorithms,

traditional models that rely on predefined particularly ML techniques, can be used to
rules and historical credit data, AI models build predictive models that identify
can process a myriad of information, patterns and correlations in historical credit
including alternative data,such as utility data, forecasting the greater likelihood of
payments and rental history, to create a loan defaults.
more comprehensive assessment of an
individual's creditworthiness, ensuring that Behavioral analysis: Many individuals
a broader range of financial behaviors is encounter hurdles in the form of limited or
considered. nonexistent credit history, making it difficult
to prove their creditworthiness due to the
AI offers unparalleled adaptability: As lack of historical data. AI algorithms can
economic conditions change and consumer analyze behavioral data sets to understand
behaviors evolve, AI-powered models can financial habits and risk propensity. By
quickly adjust and learn from new data. This monitoring real-time financial behavior, AI
continuous learning ensures that credit models can provide dynamic credit scores
scoring remains relevant and effective in that reflect current risk profiles.
ever-changing financial landscapes.
Lack of clarity in rejection reasons: A
Fraud detection: AI algorithms can detect lack of transparency in rejection reasons
fraudulent behavior by identifying leaves applicants in the dark, making it
anomalies and suspicious patterns in credit difficult for them to address the root cause
applications and transaction data. and enhance their creditworthiness for
future applications.
By harnessing the power of artificial
intelligence, lenders can make more informed
lending decisions, expand access to credit, and
better serve consumers (especially those with
limited credit history). However, to mitigate
potential biases and ensure consumer trust, it's
crucial to ensure transparency, fairness, and
regulatory compliance when deploying
artificial intelligence in credit scoring.

The most common objections from banks to

not using AI in credit scoring are transparency
and explainability in credit decisions. The
inherent complexity of some AI models,
especially deep learning algorithms, may lead
to challenges in providing clear explanations
Gen AI has the potential to revolutionize credit
for credit decisions. Fortunately, the
scoring and assessment with its ability to
transparency and interpretability of AI models
create synthetic data and understand intricate
have seen significant advancements.
patterns, offering a more nuanced, adaptive,
Techniques like SHapley Additive exPlanations
and predictive approach. Gen AI’s capability
(SHAP) values and Local Interpretable
to synthesize diverse data sets addresses one
Model-Agnostic Explanations (LIME) plots and
of the key limitations of traditional credit
several other advancements in the domain of
scoring—the reliance on historical credit data.
Explainable AI (XAI) now allow us to
By creating synthetic data that mirrors
understand how the model arrives at specific
real-world financial behaviors, gen AI models
credit decisions. This not only enhances trust in
enable a more inclusive assessment of
the credit scoring process but also addresses
creditworthiness.
the common critique that AI models are "black
boxes." Adaptability plays a crucial role in navigating
the dynamic nature of economic conditions
and changing consumer behaviors. Unlike
traditional models that struggle to adjust to
unforeseen disruptions, gen AI’s ability to
continuously learn and adapt ensures that
credit scoring remains effective in real-time,
offering a more resilient and responsive tool
for assessing credit risk.

In addition to its predictive prowess, gen AI

can contribute to transparency and
interpretability in credit scoring. Models can
generate explanations for their decisions,
providing clearer insights into credit
assessments, and enhancing trust among
consumers, regulators, and financial
institutions.
 Enhance credit applications and Below is an architectural diagram of the
scoring with MongoDB in the data processing pipeline for the predicting
following key aspects of the process: probability of delinquency and credit
scoring. The data pipeline for credit scoring
Simplify data capture and processing: a customer involves the following steps:
streamlining this process by eliminating data collection, data processing, risk profile
redundant requests and leveraging existing generation, model development, data
data could significantly enhance the user transformation, and decision collection.
experience. Leveraging JSON for online
credit application forms simplifies the data Explain the credit application

capture process and also the performance declination: when it comes to credit

in data processing. JSON's structured data application declination, understanding the

representation proves highly conducive for reasons behind it is crucial. Let’s explore

organizing the multifaceted information how MongoDB and large language models

within credit applications, encompassing (LLMs) can shed light on XGBoost model

personal, financial, and employment details. predictions (the model used in this tutorial).
The risk profiling ML pipeline employed
Enhance credit scoring with AI: provides a probability score that defines the
Leveraging MongoDB’s developer data risk associated with the profile for product
platform — an integrated suite of data recommendation. This message is
services centered around a cloud database communicated back to the user in a
— we can create a comprehensive templatized manner where only the final
customer/user banking profile by status of the application is communicated
combining relevant data points. to the end user.

Figure 29: Architecture of the data processing pipeline for the predicting probability of
delinquency and credit scoring
 Below you will see the architecture
explaining credit scoring using an LLM.

Figure 30: Architecture of the credit application declination

Recommend alternative credit products: already engaged in the process. Below is an

If the credit product applicant is declined, example of a data architecture that is used
the credit institution should still try to to achieve this.
cross-sell to the customer with a relevant
product that meets their needs as they are

Figure 31: Architecture of the alternative credit product recommendations

Learn how with this credit card Credit card application with generative AI tutorial
 Assessing Business Loan
Risks

Business loans are a cornerstone of banking For instance, ChatGPT-4.0 can be prompted to
operations, providing significant benefits to perform a flood risk assessment. However, due
both financial institutions and the broader to insufficient flood data, it first indicated the
economy. In 2023, the value of commercial location was in a low flood risk zone but later
and industrial loans at commercial banks in claimed it had significant flooding. This
the United States reached nearly 2.8 trillion inconsistency can confuse users and lead to
U.S. dollars. However, these loans involve incorrect risk assessments.
significant risks, including credit and business
risks.

Importance of Business Plans

A comprehensive business plan is essential in

securing business loans, providing detailed
insights into a borrower’s strategy, market
position, operational plans, and financial
forecasts. However, the complexity and length The full risk assessment is more detailed

of these plans pose challenges for loan

officers, including time constraints and the
need for specialized knowledge to interpret key
metrics and risk factors accurately.

Generative AI in Risk Assessment

Generative AI can streamline the analysis of

business plans by extracting essential
information and highlighting key risks. This
includes using gen AI-powered chatbots that
simulate discussions with loan officers,
providing real-time risk assessments based on
the input data.

Figure 32: Examples of how ChatGPT-4.0 could

respond when asked to assess the risk of a
business loan and how it responded to flood risk as
a factor
US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 Retrieval-augmented generation (RAG) and Demonstration and Results
MongoDB
Let's examine several figures from a
While gen AI offers significant potential, its generative AI RAG demo built using LLama 3
effectiveness in loan officer workflows is on the Fireworks.AI platform, with MongoDB
hindered by the need for constant prompting as an operational data store for geospatial
and context augmentation. This manual queries. In this demo, users are asked for a
process is time-consuming and often requires business location using a map (where they can
specialized skills. To overcome these challenges either choose one of the defaults from a
and enhance data accuracy, we propose a dropdown list or use a geolocation pin to pick
retrieval-augmented generation (RAG) a location on the map) and a short business
approach utilizing MongoDB's multimodal purpose and description (to emulate the idea
capabilities. By incorporating geospatial data of a business plan).
and other external datasets, RAG can provide
a more comprehensive risk analysis. For
instance, MongoDB's geospatial queries can
identify historical flood occurrences near a
business location, enriching AI-generated
insights with crucial context.

Figure 33: Choosing a location on the map and writing a brief plan description
 Upon submission, the AI performs a RAG
analysis. Using prompt engineering, it delivers
a simplified business analysis considering the
location and flood risk data from external
sources. McKinsey's survey found that 65% of
respondents report their organizations are
regularly using generative AI, including RAG, in
at least one business function. This
demonstrates the increasing adoption and
integration of RAG within business processes.

Figure 34: Loan risk response using RAG

Clicking the "Pin" icon next to the location bar

reveals sample flood locations near the
selected business site. The pins mark flood
locations, and a blue circle shows a 5km radius
where flood data is queried.

Figure 35: Flood locations displayed with pins

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 RAG and multimodal data: a robust
approach to loan risk assessment

Integrating gen AI with a multimodal

developer data platform like MongoDB Atlas
offers a robust solution for detailed and
accurate business loan risk assessments. The
iterative nature of the RAG process ensures
continuous learning and improvement,
minimizing risks like data hallucinations. As AI
models become increasingly multimodal,
leveraging comprehensive data platforms will
be crucial for maximizing their potential.

Solution: Better Business Loans with

MongoDB and Generative AI.

Figure 36: RAG data flow architecture diagram

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 Emerging use cases
for AI in payments

A lack of developer capacity is one of the are several areas in which banks can apply AI
biggest challenges for banks when delivering to unlock new revenue streams and efficiency
payment product innovation. Banks believe the gains. The image below provides a high-level
product enhancements they could not deliver view on eight of the principal themes and
in the past two years due to resource areas. This is not an exhaustive view but does
constraints would have supported a 5.3% demonstrate the depth and breadth of current
growth in payments revenues. With this in opportunities. In each example, there are
mind and the revolutionary transformation already banks that have begun to bring
with the integration of AI, it is imperative to services or enhancements to market using AI
consider how to free up developer resources to technologies or are otherwise experimenting
make the most of these opportunities. There with the technology.

Figure 37: Benefits of payment processing optimization

Below are some areas according to tech value-added services making the most of
analyst Celent in which AI can be leveraged payments data. Analysis of payables,
across financial institutions in front, middle receivables, and balance information in
and back office operations. aggregate can be used to deliver a range of
different insights and service improvements
Front office for new and richer insights: in real-time. Below are three areas where
AI has huge potential to support
financial institutions can use AI to enhance ● Product and service
their customer-facing services: enhancements: Personalizing
services is a huge area AI can help.
● Cash flow analysis and
forecasting: Using the information Cross-functional capabilities to
in payment messages to provide increasing agility and efficiency
granular and real-time insights into AI technologies can also be used to support
the cash position for a client. several cross-functional use cases within a
● Working capital optimization: bank:
Using AI in conjunction with
payments data can support a range ● Using gen AI to support code
of use cases around liquidity and generation: This is a huge
management with insights and opportunity for institutions to
recommendations. increase developer efficiency.
● Payment insights: Gen AI can ● Risk, compliance, and reporting:
support the delivery of payment and Technologies such as machine
account insights to customers. learning, NLP, computer vision, and
gen AI can all be deployed in
Middle and back office for workflow and various combinations to support
process improvements: processes such as risk management
AI and advanced analytics are already and regulatory reporting.
commonplace across large areas of middle
and back-office operations for payment
processing, but many potential
Dive into Celent’s report to learn
opportunities for further enhancement still
more about the use cases above
exist:
and understand how you can
● Payment processing harnessing the benefits of AI in
optimization: One use case for payments.
machine learning is to further
automate the payments repair
process. AI technologies to improve
transaction routing is another
important opportunity. A third would
be in message translation or
conversion between older standards
and ISO formats, e.g. to translate
unstructured name and address in
ISO 8583 messages to structured
ISO 20022 messages.
● Risk and fraud: As with the use
case for payments repair, applying
AI technologies in transaction
screening can reduce the need for
manual intervention.
 Regulatory Compliance and
Code Change Assistance

The regulatory landscape for banks has grown AI has the capacity to revolutionize
increasingly complex, demanding significant compliance by automating the translation of
resources for the implementation of numerous regulatory texts into actionable data
regulations. Traditionally, adapting to new requirements and validating compliance
regulations has required the manual through intelligent analysis. This approach is
translation of legal text into code, provisioning not without its challenges, as AI-based
of data, and thorough quality control—a systems may produce non-deterministic
process that is both costly and outcomes and unexpected errors. However, the
time-consuming, often leading to incomplete ability to rapidly adapt to new regulations and
or insufficient compliance. For instance, to provide detailed records of compliance
comply with the Basel III, Dodd-Frank, PSD2, processes can significantly enhance regulatory
or various other regulations, developers must adherence.
undertake extensive coding changes to
accommodate the requirements laid out in
thousands of pages of documentation.

Recent research highlights that generative in identifying gaps and interpreting new
AI not only identifies regulatory obligations regulatory requirements improves. Eventually,
and gaps but also provides accurate code generative AI may even write new controls,
recommendations. For example, generative driving greater efficiency.
AI can help a bank ingest all applicable
regulatory requirements and internal This seamless process is facilitated by using a
controls, quickly alerting it to any multimodal database like MongoDB, which
discrepancies. By harnessing its addresses the common issue of documents
comprehensive and sophisticated being scattered and lacking proper metadata
capabilities, generative AI can significantly and documentation. MongoDB offers several
reduce manual effort and minimize errors, key differentiating features:
all while strictly adhering to factual data.
● Converged operational and AI data
It then determines an organization’s storage: MongoDB allows
obligations and evaluates its compliance. If organizations to keep all their
operational, unstructured, and vector
obligations are unmet, AI maps them to
data in one place, eliminating data
specific controls or generates new ones. As
silos and making it easier to access
AI models continue to train, their accuracy and manage.
 ● Efficient data handling: MongoDB 2. Translate to controls: Convert these
simplifies working with data at scale. summaries into actionable controls. In this
Centralizing data reduces manual context, control is a specific measure often
effort, leading to quicker and more implemented through code that ensures
accurate compliance processes. compliance with regulatory requirements by
● Enhanced search and generative AI automating necessary actions, such as
capabilities: MongoDB facilitates easy enforcing multi-factor authentication (MFA) for
searching through documents and certain transactions.
metadata, crucial for quickly
identifying regulatory requirements 3. Convert into executable code:
and checking them against existing
● Check the controls against the
controls. Utilizing RAG capabilities with
organization's existing internal
MongoDB provides better context to
controls, identify gaps, and create new
LLMs for generating code suggestions.
controls where needed.
Below is an example of a conceptual workflow ● Convert controls into executable code.
using MongoDB for code generation:

1. Regulatory requirements: Gather relevant

regulatory requirements, determine the
specific obligations from these regulations,
and condense the regulatory text into concise
summaries.

Figure 38: Automated compliance code adaptation with AI

 In the financial services section, we Step 2: Control
emphasized the importance of regulatory Ensure that any electronic payment initiated
compliance by referencing several key by the payer requires multi-factor
regulations such as AML, Basel III, PCI, and authentication (MFA), unless an exemption
applies.
GDPR. To demonstrate the practical
application of code generation, we used
Step 3: Code suggestion
PSD2 as an example and engaged ChatGPT
Using MongoDB and generative AI, the
to see if this approach is feasible. Below, you
process might look like this:
will find the ChatGPT prompts and
responses. 1. Identify payment transactions:
Store all electronic payment
Step 1: Real PSD2 clause transactions in MongoDB.
Banks must apply strong customer 2. Check transaction type: Create a
authentication (SCA) for electronic control to check if the transaction is
payments initiated by the payer, unless an initiated by the payer.
exemption applies. 3. Implement MFA: Generate code to
enforce multi-factor authentication
(MFA) for applicable transactions.

The full example is more detailed

Figure 39: ChatGPT 4.0 Example of PSD2 regulation converted into code

Below is a simple example of what the

generated code for MFA might look like.

Figure 40: MongoDB setup for multi-factor Figure 41: Transaction handling and strong
authentication customer authentication (SCA) enforcement
 Other Notable Use Cases
Below are a few more examples of where AI These examples highlight several avenues for
can be leveraged in financial operations: integrating AI within financial institutions.
Embracing AI in financial applications
Risk modeling: AI can create synthetic promises enhanced risk management,
scenarios and data that can be used to operational efficiency, and superior customer
stress test financial systems and models experiences.

Algorithmic trading: AI algorithms can Therefore, it is essential for financial

analyze public market data and execute institutions to grasp the profound
trades at speed, optimizing trading technological implications, scale, and
strategies. intricacies associated with AI, particularly in
crafting a generative AI strategy. Adopting a
Generating financial reports: AI can
strategic and holistic approach that
analyze financial data including transactions,
addresses technological, data, ethical, and
invoices, and account statements to
organizational dimensions is imperative for
automate the process of generating reports.
navigating this transformative landscape
By using AI and ML techniques, relevant
effectively.
information can be extracted where required.

Bendigo and Adelaide Bank is a leading Australian financial

institution with over two million customers and 7,000 employees.
Bendigo and Adelaide Bank uses By prioritizing digital transformation, the bank successfully

gen AI and MongoDB to modernize migrated 32 banking applications to AWS in just 30 days,
leveraging MongoDB to accelerate development and deliver
applications
enhanced services to its customer base.

Contact Information

Wei You Pan Luis Pazmino Diaz

Global Director, Financial Principal, Financial
Services Industry Services Industry
[email protected] [email protected]

Paul Rogers
Principal, Financial
Services Industry
[email protected]

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
 Atlas for Industries

Insurance

With its ability to

streamline processes,
enhance decision-making,
and improve customer
experience with far less
time, resources, and staff
than traditional IT
systems, AI offers insurers
great promise.
Architecting with AI for the Insurance Industry

Delivering business value Claim as an example

When architecting for AI-augmented Common organizational goals for an insurer
application workflows, it’s important to keep may include:
your overall business objectives in mind. How
will what you do help the insurance business ● Operational excellence: Emphasizing
you support? What problem are you solving? efficiency and effectiveness in
operations to maximize returns and
IT portfolio managers who may be driving reduce waste
platform consolidation and modernization ● Customer centricity: Prioritizing
agendas inside of your organization may be investments that improve customer
approaching the space from a technology-first satisfaction and engagement
lens, evaluating one platform vs. another in
terms of technical capabilities and cost. Data How well your organization can process and
science and analytics folks may be wishing to resolve claims, for example, directly impacts
test or prove data hypotheses. Business the objectives above. Achieving this is directly
product owners may have priorities in terms of coupled to how quickly, efficiently, and
raising the bar when it comes to your business accurately we can process the data found
applications and their features, hoping to within claim-handling application workflows.
augment their data processing capabilities
with the addition of AI. Claim handling provides us great opportunities
to exploit AI in order to accelerate data
For you, as the architect or CxO, your charge is processing hotspots, so that the organization
to help the organization leverage AI to make can return meaningful value from the
meaningful strides in terms of business technology investment.
outcomes, and that means navigating a
number of dimensions, not limited to: So what data in a claim-handling workflow is
hard to work with? Unstructured data sources
● Business objectives: A clear such as damage photos, accident forms and
understanding of the organization's reports, claim handler notes, traffic camera
goals videos, and claim-handling guidelines and
● Data and workflows: Knowledge of recommendations. Are any or all of these data
business processes and supporting sources cumbersome to work with for your
data employees? Do they need to open and read
● AI application: Applying AI to forms, examine and interpret images, and
streamline data-intensive tasks distill and write case-file notes before your
claims can move forward in the handling
Our end goal is to be able to gather, lifecycle? If a catastrophic event strikes, do
understand, interact with, and generate data you have the staff to handle a massive, sudden
faster by applying the right AI capabilities to influx of claims, all with these types of
the right points in the organization. unstructured data?
Augmenting with AI Architecting for applications
Architects are the ones tasked with deciding Your software delivery teams and the
what type of AI is best suited to solve for applications they support may be divided up
insurance data processing needs. by agile delivery domains. It’s inside of those
domains and applications that you’ll need to
Machine learning models trained on historical apply AI in order to effectively move the
data can be used to make predictions and needle on organizational and process
decisions within a business workflow, in lieu of outcomes. In short, your AI belongs in your
a person. applications.

Generative AI and large language models, The data that supports those applications
however, give us core NLP (natural language resides in operational data stores. If we want
processing) capabilities that are particularly both our applications and our AI to work with
well-suited to augmenting data processing real-time data, it should be accessible within
abilities. These capabilities, as applied to the the same underlying data store. What serves
claim-handling workflow, include: our apps should also serve our AI.

Entity extraction to help interrogate and Managing operational structured and

retrieve relevant information from an unstructured data:
unstructured source such as a PDF guideline, Your application architecture needs to provide
or a large body of text found within an a way to store, serve, and update structured
accident form. data as part of your workflow and integrate
the unstructured data. This may come in the
Text and image classification enables claim form of both the raw data (PDF’s, images,
handlers to determine types of damage or notes), and the vector-encoding representation
characteristics found on a damage photo. of it. You’ll want the vector-encoding data to
be as close as possible to the structured data
Text summarization speeds up the synthesis of that already may exist in your application.
large bodies of text, or across many sources. Several reasons why include:

Text generation helps generate case files and ● Performance: Providing

give succinct instruction to workers based on application-level SLA (service level
larger bodies of guidelines, for example. agreements) utilizing efficient server
compute
Interactive chat makes the previous or ● Security: Providing consistent app-tier
additional existing information available more level security controls to determine who
quickly to both your employees and your can access what data.
customers. ● Ease of application delivery and
maintenance: The fewer components
and dependencies, the lower the
complexity and cost of building,
deploying, and maintaining effective
software solutions that leverage AI.
Applying AI to the claim-handling workflow
and delivering better business outcomes

Figure 42: AI use cases aligned to the claim-handling workflow

The above diagram shows practical use of core for example, these capabilities can reduce
NLP capabilities that include text and image data hotspots, resulting in lower processing
classification, interactive chat, entity times and costs, and improved customer
extraction, and text summarization. When experiences.
applied to a claim-handling workflow,
Before AI can transform our organizations,
we first have to bring it into our applications

Figure 43: AI Maturity levels from undefined to transformational

The above diagram describes the different AI-powered apps and experiences: The
stages of AI adoption within enterprises, from interface and experiences that end-users
early interest to a pervasive and structural interact with, and app frameworks that
integration into processes and decisions. Many simplify the process of building AI experiences.
organizations struggle with moving out of
Level 2, analytics experimentation, to Level 3, Foundational models are extremely powerful,
deploying AI features within business but being trained on public datasets, they lack
applications, in order to deliver meaningful the domain knowledge and data context
business value and outcomes. needed to adequately support enterprise
applications.
The three layers of generative AI
Gen AI applications can be separated into This is where the second layer comes into play:
three major layers: the data and tooling that will glue everything
together and enable generative AI-enhanced
Compute and AI models: Underlying applications to be fully operational, moving
processing power plus foundational and your organization from the experimentation
embedding models. level of Level 2 into full-fledged production
apps of Level 3 maturity.
Tooling to fine tune and build applications:
Tools that provide foundational models with
context by feeding them proprietary data.
 MongoDB’s operational database stores
AI-powered
proprietary structured and vector data, and
apps and experiences
makes it available to LLMs when an
application makes a request, effectively Data and tooling to use the
providing the context necessary for a model
foundational model to answer questions
Compute and AI models
beyond its initial knowledge boundary.

Building applications that leverage AI

Figure 44: Applications with AI

The above diagram illustrates in greater detail processed. Part of the processing includes the
the three layers of gen AI, showing the vectorization and subsequent persistence of
structured data flowing from the application those vectors in the operational data store,
into the operational database, raw where it can be readily accessed by the
unstructured data managed in object storage, application.
accessible by applications where it can be
 In an inherently information-driven industry,
insurance companies ingest, analyze, and
process massive amounts of data. Whether
it’s agents and brokers selling more policies,
underwriters adequately pricing, renewing
and steering product portfolios, claim
handlers adjudicating claims, or service
representatives providing assurance and
support, data is at the heart of it all.

Given the volumes of data, and the amount IT investments. With its ability to streamline
of decision-making that needs to occur processes, enhance decision-making, and
based on it, insurance companies have a improve customer experiences with far less
myriad of technologies and IT support staff time, resources, and staff than traditional IT
within their technology investment portfolios. systems, AI offers insurers great promise.
It’s no surprise that AI is at the top of the list
when it comes to current or prospective

Few roles within insurance are as important

Underwriting & as that of the underwriters who strike the

Risk Management right balance between profit and risk, bring

real-world variables to the actuarial models
at the heart of the insurer, and help steer
product portfolios, markets, pricing, and
coverages. Achieving equilibrium between
exposures and premiums means constantly
gathering and analyzing information from a
myriad of sources to build a risk profile
sufficient and detailed enough to make
effective policy decisions.
While many well-established insurers have Advanced analytics
access to a wealth of their own underwriting Traditional IT systems are slow to respond
and claims data, integrating newer and to changing formats and requirements
real-time sources of information, keeping up surrounding data retrieval. The burden falls
with regulatory changes, and modeling out on the underwriter to summarize data and
what-if risk scenarios still involve significant turn that into information and insight. Large
manual effort. language models are now being leveraged to
help speed up the process of wrangling data
Perhaps the single greatest advantage of AI sources and summarizing the results, helping
will be its ability to quickly analyze more underwriting teams make quicker decisions
information with fewer people and resources. from that data.
The long-term impact will likely be profound,
and there is tremendous promise within
underwriting. Workload and triage assistance
AI models are mitigating seasonal demands,
market shifts, and even staff availability that
impact the workload and productivity of
underwriting teams, saving underwriting time
for the high-value accounts and customers
where their expertise is truly needed. Amid
high volumes for new and renewal
underwriting, traditional AI models can help
classify and triage risk, sending very low-risk
policies to touchless automated workflows,
low to moderate risk policies to trained
service center staff, and high-risk and high
value accounts to dedicated underwriters.

Decision-making support
Determining if a quoted rate needs adjustment
can take significant time and manual effort. So
can preparing and issuing renewals of existing
policies, another large portion of the
underwriters’ day-to-day responsibilities.
Automated underwriting workflows leveraging
AI are being employed to analyze and classify
risk with far less manual effort. This frees up
significant time and intellectual capital for the
underwriter.
Vast amounts of data analyzed by For every major business entity found within
underwriters are kept on the underwriters the underwriting process, such as broker,
desktop rather than IT-managed databases. policy, account, and claim, there is a wealth
MongoDB offers an unparalleled ability to of unstructured data sources, waiting to be
store data from a vast amount of sources leveraged by generative AI. MongoDB offers
and formats, and respond quickly to insurers a platform that consolidates
requests to ingest new data. As data and complex data from legacy systems, supports
requirements change, the document model modern applications, and extends those
allows insurers to simply add more data and same data assets to AI-augmented
fields without the costly change-cycle workflows. By eliminating the need for niche
associated with databases that rely on databases for these AI-specific workloads,
single, fixed structures. MongoDB reduces technology evaluation
and on-boarding time, development time,
and developer friction.

Claim Efficient claim processing is critical for an

insurer. Timely resolution of a claim and good
Processing communication and information transparency
throughout the process are key to maintaining
positive relationships and customer
satisfaction. In addition, insurers are on the
hook to pay and process claims according to
jurisdictional regulations and requirements,
which may include penalties for failing to
comply with specific timelines and stipulations.
In order to process a claim accurately, a Claim expediter
wealth of information is needed. A typical Using vector-embeddings from photo, text,
automobile accident may include not only and voice sources, insurers are now able to
verbal and written descriptions from decorate inbound claims with richer and more
claimants and damage appraisers but also insightful metadata so that they can more
unstructured content from police reports, quickly classify, triage, and route work. In
traffic and vehicle dashboard cameras, addition, real-time insight into workload and
photos, and even vehicle telemetry data. staff skills and availability is allowing insurers
Aligning the right technology and the right to be even more prescriptive when it comes to
amount of your workforce in either single- or work assignments, driving towards higher
multi-claimant scenarios is crucial to output and higher customer satisfaction.
meeting the high demands of claim
processing.

Litigation assistance
Claim details are not always black and white,
parties do not always act in good faith, and
Taming the flood of data insurers expend significant resources in the
AI is helping insurers make sense of a trove pursuit of resolving matters. AI is helping
of data faster and in real time. From natural insurers drive to resolution faster and even
language processing to image classification avoid litigation and subrogation altogether,
and vector embedding, all the pieces of the thanks to its ability to help analyze more data
puzzle are now on the board for insurers to more effectively and in less time.
make a generational leap forward when it
comes to transforming their IT systems and
business workflows for faster information
processing.
Risk prevention
Many insurers provide risk-assessment services
to customers using drones, sensors, or
cameras to capture and analyze data. This
data offers the promise of preventing losses
Claims experience
altogether for customers and lowering
Generating accurate impact assessments for
exposures, liability, and expenses for the
catastrophic events in a timely fashion in order
insurer. This is possible thanks to a
to inform the market of your exposure can now
combination of vector embedding and
be done with far less time, and with far more
traditional and generative AI models.
accuracy, by cross-referencing real-time and
historical claims experience data, thanks to the
power of generative AI and vector-
embedding of unstructured data.
 Customer Accessing information consistently during a
customer service interaction, and expecting
Experience the representative to quickly interpret it, are
perennial challenges with any customer
service desk. Add in the volume, variety,
and complexity of information within
insurance and it’s easy to understand why
many insurers are investing heavily in
transformation of their customer experience
call center systems and processes.

24/7 virtual assistance Customer profiles

As with many AI-based chat agents, the Every interaction is an opportunity to learn
advantage virtual assistants is that they can more about your customers. Technologies such
free up your call center staff to work on as voice-to-text streaming, vector embedding,
more complex and high-touch cases. and generative AI help insurers build out a
Handling routine inquiries can now extend to more robust social profile of their customers in
more complex scenarios than ever thanks to near real-time.
the power of vector-embedded content and
large language models.

Claims assistance Real-time fraud detection

Generative AI can deliver specific According to estimates from the Coalition
claim-handling guidelines to claim-handling Against Insurance Fraud, the U.S. insurance
staff in real time, while traditional ML models industry lost over $308 billion to fraud in 2022.
can interrogate real-time streams of With vector-embedding of unstructured data
collected information to alert either the sources, semantic and similarity searches
customer or the claim-handler to issues with across both vector and structured metadata,
quality, content, or compliance. AI and traditional machine learning models,
capabilities allow insurers to process more insurers can detect and prevent fraud in ways
claims faster and significantly reduce errors that were simply not ever before possible.
or incomplete information.
 Domain- Insurance enterprises all over the world
have embraced domain-driven design in
Driven AI addition to aligned software delivery teams
and the applications they support by core
processing domains. With a shift to
microservices and event-streaming, AI
capabilities can now supercharge this
architecture and accelerate the ability to
interact with and serve up real-time data.
Domain-based operational data stores help
segment and enable stronger data lineage,
data quality, and data governance for more
truthful AI interactions.

Working together: Applications,

data & AI customer
applications

Core insurance business capabilities

data AI
organized and deployed by domain and
domain capabilities augmented by AI:

● Domain- and task-specific AI, submission

enhancing key application workflows applications
and data processing hotspots policy
● Operational and vector data stored data AI
applications
as close to applications as possible
for: data AI
○ Real-time context
○ Performance
○ Security & privacy claim
○ Domain agility applications
● APIs and events for efficient
processing within and across data AI
domains.
Examples of
Domain-Specific AI

Ferret.ai builds trust with AI &

“Through Atlas, itʼs available as a
MongoDB Atlas, slashes costs 30%
fully managed service with best
Ferret.ai helps companies and individuals practices baked in. That frees
build trust by providing real-time insights on developers and data scientists from
people and businesses using AI. Powered by
MongoDB Atlas, Ferret analyzes vast
the responsibilities of running a
amounts of public data to identify risks and database so they can focus their
opportunities, offering a comprehensive efforts on app and AI innovation”
view of relationships.
Learn more
Al Basseri
CTO and head of Engineering at Ferret.ai

VISO TRUST: Accelerating

and enhances overall cybersecurity posture.
third-party risk assessment with AI The platform uses sophisticated models and
retrieval-augmented generation (RAG)
VISO TRUST is an AI-powered third-party
techniques to provide highly accurate and
cyber risk and trust platform that enables
contextually relevant intelligence. This not
any company to access actionable vendor
only accelerates decision-making but also
security information in minutes. By
ensures that companies are equipped with
leveraging MongoDB Atlas for efficient data
the most precise risk evaluations available.
storage and retrieval, VISO TRUST's
Assessments generated from analysis are
advanced AI models can process vast
80% faster than manual methods with no
amounts of security information, delivering
loss in accuracy
actionable insights in minutes. This
streamlined approach significantly reduces Learn more
assessment time
 Solution Library The Solutions Library is a curated collection
of tailored solutions, offering ready-made
References templates, best practices, and detailed
guides to help developers quickly start
projects. With industry-specific examples
and sample code, the library enables
insurance companies to efficiently
implement common patterns, leverage
MongoDB's features effectively, and ensure
compliance with industry standards. Here
are some examples of solutions for the
insurance industry.

Claim management using LLMs and vector search for RAG

By converting claim data into vector insights and context, optimizing claim
embeddings, MongoDB's Atlas Vector Search processing. This combined approach
accelerates information retrieval, making it enhances accuracy, efficiency, and overall
quicker and easier to find relevant details. claims management.
Large language models (LLMs) then analyze
these embeddings to extract valuable Learn more

Figure 45: Atlas Vector Search provides answers to user questions about insurance claims, including
calculations and detailed claim examples
AI-enhanced claim adjustment for auto insurance
By leveraging AI and vector image search, Centralized data management in MongoDB
this solution automates auto insurance Atlas enhances efficiency and consistency
claim adjustments. Accident photos are throughout the claims process.
compared to a database of past claims,
significantly accelerating damage estimates. Learn more

Figure 46: An image similarity query is performed, and the 5 top similar images are returned

Build a PDF search application with vector search and LLMs

PDFs are hard to search, making it tough for Superduper, allowing users to quickly retrieve
insurance workers to find what they need information and streamline insurance work.
quickly. We’re solving this by converting PDFs
into a searchable format using tools like Learn more

Figure 47: PDF chunking, embedding creation, and storage, orchestrated with Superduper
Other Notable Use Cases

Predictive analytics Predictive maintenance

AI-powered predictive analytics can for property insurance
anticipate customer needs, preferences,
AI-powered predictive maintenance solutions,
and behaviors based on historical data
leveraging IoT sensors installed in buildings and
and trends. By leveraging predictive models,
infrastructure, are used in property insurance to
insurers can identify at-risk customers,
prevent losses and minimize damage to insured
anticipate churn, and proactively engage
properties.
with customers to prevent issues and
enhance satisfaction.

Crop insurance and Usage-based insurance (UBI) for

precision farming commercial fleets
AI is being used in agricultural insurance to AI-enabled telematics devices installed in
assess crop health, predict yields, and commercial vehicles collect data on driving
mitigate risks associated with weather behavior, including speed, acceleration,
events and crop diseases, which helps braking, and location. Machine learning
insurers offer more accurate and tailored algorithms analyze this data to assess risk
crop insurance products to farmers. and determine insurance premiums for
commercial fleets to help promote safer
driving practices, reduce accidents, and
lower insurance costs for businesses.

Contact
Information
Jeff Needham
Insurance Industry
Solutions Principal
[email protected]

Visit MongoDB Atlas for Insurance

 Atlas for Industries

Healthcare and
Life Sciences
AI is transforming
healthcare by enhancing
decision-making, improving
patient experiences, and
boosting operational
efficiency.

In life sciences, AI
accelerates research, drives
drug discovery, and
personalizes treatments,
leading to breakthroughs in
innovation and tailored
medicine.
Transformative Potential of Generative AI in
Healthcare and Life Sciences

The healthcare and life sciences sectors have Additionally, 57% of healthcare providers
witnessed a significant surge in the application report that excessive documentation
of artificial intelligence (AI) and machine contributes to clinician burnout. This
learning (ML) over recent years. These administrative overload not only affects the
technologies have been increasingly integrated well-being of healthcare providers but also has
into various aspects of healthcare, from a negative impact on patient care, with two
diagnostics and treatment planning to out of three patients reporting a lack of
operational efficiencies and patient empathy from their healthcare providers due
engagement. The acceleration of AI and ML to these constraints.
adoption is evident in the exponential growth
of FDA-approved algorithms, reflecting a The solution: Leveraging gen AI to
broader acceptance and reliance on these alleviate administrative burden
technologies to enhance healthcare delivery.
Gen AI has the potential to dramatically
reduce the administrative workload on
However, the advent of generative AI marks a
healthcare professionals, allowing them to
new frontier in this evolution, bringing the
focus more on patient care. Technologies such
potential to disrupt and revolutionize even
as ambient listening, combined with advanced
more areas of the industry. Unlike traditional
patient records, could automate much of the
AI, which primarily focuses on analyzing and
documentation process, reducing the time
predicting, gen AI can autonomously read and
clinicians spend on EHRs. By harnessing these
interpret multimodal data, and generate new
capabilities, healthcare providers can reclaim
content, offering unprecedented possibilities in
valuable time, improve the quality of patient
automation, personalized medicine, drug
interactions, and ultimately enhance overall
discovery, and beyond.
patient outcomes.

The challenge: Administrative This technology could also facilitate the

burden and clinician burnout summarization of patient records and present
Healthcare professionals have long been the them in visually rich interfaces tailored to the
cornerstone of patient care, responsible for specific needs of each organization, specialty,
inputting vast amounts of information into and even individual healthcare providers. Gen
electronic health records (EHRs) with the AI can also assist in the documentation and
promise that it would lead to improved patient analysis of insights from medical imaging and
outcomes. However, many clinicians feel they lab results, as well as automate prior
have yet to see the tangible benefits of these authorization processes to enhance efficiency
systems. Studies reveal that clinicians spend for both providers and payers. These
twice as much time on administrative tasks as advancements would significantly lighten the
they do in direct contact with patients. workload and reduce the burden on clinicians,
allowing them to dedicate more time and
energy to patient care.
Transforming patient communication study reports (CSRs) and other essential
and engagement documents. By streamlining these processes,
Gen AI is not only transformative for gen AI can reduce the time required to bring
healthcare providers but also holds promise for new medications to market, ultimately
enhancing patient communication and benefiting patients by providing quicker
engagement. By offering personalized and access to new treatments.
customized communication strategies, gen AI
In the post-marketing phase, AI can enhance
can help create more meaningful patient
surveillance efforts by monitoring real-world
interactions, build customized patient journeys,
data for adverse effects and other key
and promote preventive care. This increased
indicators, ensuring that medications remain
engagement could lead to better patient
safe and effective after they have been
outcomes and higher satisfaction rates, as
approved.
patients feel more connected and understood
by their healthcare providers.

Impact on life sciences: Accelerating

the medication lifecycle Prioritization
In the life sciences sector, gen AI is poised to
impact the entire medication lifecycle, from With the rapid progression of AI, decision
discovery to post-marketing surveillance. In makers are now faced with the challenge of
the realm of basic research, gen AI has shown prioritizing these numerous opportunities. The
success in simulating new molecules and possibilities are vast, but the path forward
interactions, providing researchers with novel requires careful consideration and strategic
ways to interact with vast amounts of scientific planning. Keep reading to explore how industry
literature. This could accelerate the drug leaders are already applying these
discovery process, leading to faster groundbreaking technologies and discovering
development of new therapies. offering insights that can help navigate this
complex and exciting landscape.
During clinical trials, gen AI can improve the
accuracy and speed of developing clinical

Figure 48: Mckinsey: Areas believed to benefit from generative AI (% of respondents)

In healthcare, transforming data into
actionable insights is vital for enhancing
clinical outcomes and advancing patient
care. From medical professionals improving
care delivery to administrators optimizing
workflows and researchers advancing
knowledge, data is the lifeblood of the
healthcare ecosystem. Today, AI emerges
as a pivotal technology, with the potential
to enhance decision-making, improve
patient experiences, and streamline
operations—and to do so more efficiently
than traditional systems.

While they may not expect it based on past

Patient Experience interactions, patients crave a seamless

& Engagement experience with healthcare providers.

Ideally, patient data from healthcare
services, including telehealth platforms,
patient portals, wearable devices, and EHR,
can be shared securely across interoperable
channels. Unfortunately, disparate data
sources, burdensome and time-consuming
administrative work for providers, and
overly complex and bloated solution stacks
at the health system level stand in the way
of that friction-free experience.
AI can synthesize vast amounts of data and wearable data, and patient-generated health
provide actionable insights, leading to data for personalized care and better
personalized and proactive patient care, patient outcomes. For wearables and
automated administrative processes, and medical devices, MongoDB is the ideal
real-time health insights. AI technologies, underlying data platform to house time
such as ML algorithms, natural language series data, significantly cutting down on
processing (NLP), and chatbots are being storage costs while enhancing performance.
used to enhance and quantify interactions.
Additionally, AI-powered systems can On the patient care front, MongoDB can
automatically schedule appointments, send support AI-driven recommendations for
notifications, and optimize clinic schedules, personalized patient education and
all of which can reduce patient wait times. engagement based on the analysis of
AI-enabled chatbots and virtual health individual health records and engagement
assistants provide 24/7 support, offering patterns, and Atlas Vector Search can power
instant responses, medication reminders, search capabilities within patient portals,
and personalized health education. AI can allowing patients to easily find relevant
even identify trends and predict health information and resources, thereby
events, allowing for early intervention and improving self-service.
reducing adverse outcomes.

MongoDB’s flexible data model can unify

disparate data sources, providing a single
view of the patient that integrates EHRs,
Enhanced Clinical
Decision Making

Healthcare decision making is critically

dependent on the ability to aggregate,
analyze, and act on an exponentially growing
volume of data. From EHRs and imaging
studies to genomic data and wearable device
data, the challenge is not just the sheer
volume but the diversity and complexity of
data. Healthcare professionals need to
synthesize information across various
dimensions to make informed, real-time,
accurate decisions. Interoperability issues,
data silos, lack of data quality, and the
manual effort required to integrate and
interpret this data all stand in the way of
better decision-making processes.

The advent of AI technologies, particularly

NLP and LLMs, offers transformative
potential for healthcare decision making by
automating the extraction and analysis of
data from disparate sources, including
structured data in EHRs and unstructured
data in medical literature or patient notes.

By enabling the querying of databases using

natural language, clinicians can access and
integrate patient information more rapidly
and accurately, enhancing diagnostic
precision and personalizing treatment
approaches. Moreover, AI can support
real-time decision making by analyzing
streaming data from wearable devices,
alerting healthcare providers to changes in
patient conditions that require immediate
attention.
MongoDB, with its flexible data model and
powerful developer data platform, is uniquely
positioned to support the complex data needs
of healthcare decision-makers. It can
seamlessly integrate diverse data types, from
FHIR-formatted clinical data to unstructured
text and real-time sensor data, in a single
platform. By integrating MongoDB with large
language models, healthcare organizations
can create intuitive, AI-enhanced interfaces
for data retrieval and analysis. This integration
not only reduces the cognitive load on
clinicians but also enables them to access and
interpret patient data more efficiently,
focusing their efforts on patient care rather
than navigating complex data systems.
MongoDB's scalability ensures that healthcare
organizations can manage growing data MongoDB's Atlas Vector Search further
volumes efficiently, supporting the enriches decision-making processes by
implementation of AI-driven decision-support enabling semantic search across the database.
systems. These systems analyze patient data This integrated approach enables the
in real time against extensive medical application of prefilters based on extensive
knowledge bases, providing clinicians with metadata, enhancing the efficiency and
actionable insights and recommendations, relevance of search results without the need to
thereby enhancing the quality and timeliness synchronize with dedicated search engines or
of care provided. vector stores, meaning healthcare
professionals can utilize previously
undiscoverable insights, streamlining the
identification of relevant information and
patterns.
 Enhancing Medical
Imaging with
Generative AI

Generative AI offers a transformative solution including reports, and metadata from DICOM
by automating the extraction, analysis, and files. Its support for vector search capabilities
summarization of information from medical is essential for implementing retrieval-
images. By leveraging advanced NLP, augmented generation (RAG) approaches,
embeddings, and ML techniques, generative AI which enhance the accuracy and relevance of
can rapidly analyze large datasets, identify key AI-generated summaries.
insights, and generate accurate,
comprehensive summaries. This significantly Medical visual question answering (MVQA)
reduces the time radiologists spend on manual Building on the broader capabilities of
data review, allowing them to focus more on generative AI in radiology, MVQA specifically
patient care and decision-making. targets the intersection of medical imaging,
clinical reporting, and NLP. In this sub-use
For instance, AI-powered tools can be case, generative AI is utilized to not only
integrated into radiology workflows to generate summaries but also to directly
automatically generate impressions and answer clinician-specific queries regarding
diagnostic summaries based on imaging data. medical images and reports.
These tools not only streamline the reporting
process but also enhance the accuracy of Aa an example, a clinician might ask, “What
diagnostics by cross-referencing findings with are the signs of pulmonary embolism in this CT
historical data and known medical conditions. scan?” The MVQA system would analyze the CT
Moreover, chat-based AI assistants can image, cross-reference it with the clinical
provide real-time support to clinicians by report, and provide a precise answer such as,
answering queries and offering contextually “Presence of filling defects in the pulmonary
relevant information derived from patient arteries consistent with pulmonary embolism.”
records and imaging results.
This application significantly enhances the
MongoDB’s role in supporting AI comprehension of imaging reports, providing
applications concise and clinically relevant answers, which
MongoDB plays a crucial role in enabling these improves diagnostic accuracy and speeds up
AI-driven applications. By providing a flexible decision-making processes.
and scalable database architecture, MongoDB
MVQA combined with vector search
allows for the efficient storage and retrieval of
streamlines diagnostics and supports a more
vast amounts of unstructured medical data,
targeted and effective approach to analyzing
complex medical images.
 Vector embeddings for biopsy image rapidly identify and highlight suspicious tiles

analysis within the image. This process allows

pathologists to quickly pinpoint areas of
Another compelling use case involves concern, significantly enhancing the
applying vector embeddings to store and accuracy and efficiency of their reviews.
analyze predefined images with suspicious
findings, such as those from large biopsy Additionally, MongoDB’s partnership

images. Similar to text analysis, where large ecosystem, including collaborations with AI

texts are divided into smaller chunks, large pioneers, accelerates the development and

medical images can be segmented into tile deployment of these cutting-edge solutions

sets (e.g., 100x100 pixels). in the medical field.

A vector representation is generated for The following figure shows the

each tile, capturing its unique recommended architecture to implement

characteristics. By leveraging Atlas Vector this use case using MongoDB Atlas.

Search, a specialized viewer tool can

Figure 49: Gen AI in medical imaging with MongoDB Atlas for the detection of suspicious biopsies
combined with a knowledge base chatbot
By integrating generative AI into the medical
imaging workflow, healthcare providers can
achieve significant improvements in diagnostic
accuracy, reporting efficiency, and overall
patient care. The synergy between MongoDB’s
robust data management capabilities and
advanced AI technologies paves the way for a
future where healthcare professionals can
deliver faster, more reliable diagnoses,
ultimately improving patient outcomes and
reducing operational burdens.

This use case demonstrates how generative AI,

combined with MongoDB, can revolutionize the
way medical imaging data is handled in
healthcare, offering a powerful solution to
streamline processes, enhance accuracy, and
support better patient care.

In life sciences, transforming vast amounts

of data into actionable insights is critical
throughout the entire medicine
lifecycle—from basic research to market.
With the rise of precision medicine and
genomics, AI is playing a transformative role
by tailoring treatments to individual patients
based on genetic, environmental, and
lifestyle factors. Generative AI accelerates
drug discovery, analyzes genomic data for
personalized treatment pathways, and
optimizes clinical trials.
Revolutionizing Clinical Study
Reports (CSRs) with Generative AI
and MongoDB

The pharmaceutical industry faces immense required to generate CSRs from weeks to mere
pressure to expedite the regulatory approval minutes, allowing pharmaceutical companies
process for new drugs and therapies. A critical to accelerate their time-to-market for new
component of this process is the creation of drugs.
Clinical Study Reports (CSRs), which are
comprehensive documents detailing the With MongoDB Atlas, companies can leverage
methodology, execution, and results of clinical a flexible, scalable database environment that
trials. Traditionally, compiling a CSR is a supports the dynamic and varied data
labor-intensive task, often requiring several structures inherent in clinical trials. This
weeks to complete and involving flexibility is crucial for managing the diverse
multidisciplinary teams. This prolonged data types involved in CSR generation,
timeline not only delays the introduction of including text, tables, and complex statistical
potentially life-saving treatments but also data. By using generative AI models,
incurs significant costs associated with companies can automate the drafting of CSRs,
prolonged R&D cycles. producing high-quality, compliant documents
that require minimal human intervention.
Challenge
The process of generating CSRs is complex, For example, these AI models can automate
involving the integration of vast amounts of the importation and transformation of data
clinical data, including statistical outputs and tables, generate accurate narratives, and
detailed narratives. Manual methods are ensure that the final documents meet the
time-consuming and error prone, which can compliance standards required by regulatory
further delay regulatory approvals. Plus, the bodies. MongoDB’s Vector Search capabilities
need for compliance with stringent regulatory further enhance this process by enabling the
standards, such as those set by the FDA and retrieval of relevant data with high precision,
EMA, adds another layer of complexity to the which the AI uses to generate consistent and
document creation process. accurate content.

Solution: Generative AI and MongoDB

Atlas
Generative AI, integrated with MongoDB Atlas,
offers a groundbreaking solution to these
challenges by automating the CSR creation
process. This approach can reduce the time
Extended benefits to any medical writing Conclusion
The same approach described here for CSRs By integrating generative AI with MongoDB
can provide an end-to-end approach that Atlas, pharmaceutical companies can
covers a wide range of regulatory documents, transform their approach to generating clinical
including CSRs, clinical trial narratives (CTNs), study reports. This solution offers unparalleled
summary clinical safety (SCS), and summary speed, accuracy, and compliance, enabling
clinical efficacy (SCE). This comprehensive companies to bring new treatments to market
coverage ensures that companies can faster while maintaining the highest standards
automate much of their regulatory of quality and regulatory adherence. The result
submissions, reducing the risk of human error is a more efficient drug development process
and speeding up the entire process. that ultimately benefits patients by
accelerating access to innovative therapies.

This innovation is helping Novo Nordisk get

Novo Nordisk accelerates drug
new medicines to patients faster, enhancing
approval with gen AI and MongoDB both the speed and quality of their
Atlas regulatory submissions. By leveraging
MongoDB Atlas’s capabilities, NovoScribe
Novo Nordisk, a global leader in automates complex data retrieval and
healthcare, is transforming how it brings analysis, allowing the company to scale its
new medications to market using operations efficiently and securely across
generative AI and MongoDB Atlas. Known multiple cloud platforms.
for its pioneering work in diabetes care,
Novo Nordisk produces 50% of the world’s
“Weʼve reduced the time taken to
insulin and serves millions of patients
worldwide. create clinical study reports from 12
weeks to 10 minutes, with higher
NovoScribe: Revolutionizing regulatory
quality outputs and a fraction of the
submissions
team. In terms of value, each day
With the introduction of NovoScribe, the sooner a medicine gets to market
company has significantly reduced the time
can add around $15 million in
required to generate clinical study reports
(CSRs), which is a critical step in the revenue to the company.”
regulatory approval process. Learn more
Waheed Jowiya
NovoScribe, built on Amazon Bedrock, Digitalisation Strategy Lead at Novo
LangChain, and MongoDB Atlas, has
Nordisk
enabled Novo Nordisk to reduce the time to
compile CSRs from 12 weeks to just 10
minutes.
 Accelerating Drug
Discovery with
Generative AI

Current state and challenges Leveraging advanced technologies to

transform drug discovery
In the pharmaceutical industry, the drug
discovery process is a complex, The drug discovery process is poised for
resource-intensive endeavor, often involving transformation through the integration of
advanced technologies such as vector
extensive experimentation, data integration,
embeddings, SMILES notation and MongoDB
and analysis. Traditional methods, which rely
Atlas. By converting both textual information
heavily on manual research and iterative and molecular data into vector
testing, can be slow and costly, delaying the representations, researchers can unlock new
time to market for new drugs and driving up efficiencies and insights, setting the stage for
overall development expenses. more effective NLP applications in drug
discovery.
A key challenge in this domain is managing
and interpreting vast volumes of diverse At the core of this transformation is the ability
data. Researchers must sift through enormous to convert diverse data types—ranging from
datasets of molecular structures, chemical textual descriptions in scientific literature to
reactions, and historical research to identify molecular structures and chemical
promising candidates. The process requires not reactions—into vector representations. These
only identifying molecules with desirable embeddings serve as compact,
properties such as high efficacy and low high-dimensional numerical representations
toxicity but also predicting their behavior in that capture the essential properties and
complex biological systems. relationships of the data.

The iterative nature of drug discovery, ● Molecular embeddings: Molecules,

combined with these data challenges, results in including those represented by SMILES
high costs and significant time investment, notation, are transformed into vector
with no guarantee of success. This underscores embeddings that encapsulate their
the need for more efficient, data-driven structural and functional
approaches to accelerate the discovery characteristics. This allows for the
process and reduce development costs. comparison of molecules based on
their properties rather than their
chemical structures. As a result,
molecules with similar therapeutic
potential can be identified even if they
differ in their chemical makeup.
 ● Reaction embeddings: Chemical comprehensive information needed to drive
reactions can also be represented as their research forward. MongoDB Atlas
vector embeddings, capturing the supports the use of vector search technologies,
transformation of reactants to enabling efficient and accurate exploration of
products. This facilitates the search for the chemical space.
reactions that are functionally similar,
helping researchers identify alternative Enhanced discovery through Atlas Vector
Search and NLP
synthesis pathways or optimize
existing reactions.
Atlas Vector search allows researchers to
● Text embeddings: Scientific literature,
explore chemical and textual data in a way
patents, and other textual data are
that was previously impossible. When a
converted into vector embeddings that
researcher inputs a molecular structure or a
represent the semantic content of the
piece of text, the system can rapidly identify
text. This enables the system to
other molecules, reactions, or documents that
understand the context and meaning
are similar in the vector space.
behind complex chemical terms and
descriptions, making it easier to link For instance, a researcher could input the
related concepts across different vector representation of a promising molecule,
documents. and the system would return a list of similar
molecules from the database, ranked by their
By embedding all these data types into a
similarity in the vector space. This enables the
unified vector space, researchers can perform
identification of potential drug candidates that
powerful similarity searches, clustering, and
might not have been considered through
predictive modeling, significantly enhancing
traditional search methods.
their ability to discover new drug candidates.
By leveraging NLP, researchers can run
MongoDB Atlas as a developer data
complex queries in plain language, such as
platform
“What are the most recent studies on
molecules similar to this compound?” or “Show
MongoDB Atlas provides the robust
me reactions that convert this functional group
infrastructure needed to store and manage the
to another.” The NLP models interpret these
diverse data types involved in drug discovery.
queries, search the vectorized database, and
With the ability to handle both structured and
return the most relevant results, whether they
unstructured data, MongoDB Atlas supports
are molecular structures, reaction pathways, or
the integration of molecular structures, textual
related literature.
descriptions, reaction pathways, and their
corresponding vector embeddings into a
This capability drastically reduces the time and
single, scalable database.
effort required to find relevant information,
allowing researchers to focus more on analysis
This unified data platform allows for seamless
and decision making rather than data retrieval.
querying and retrieval of data. Researchers
can access molecular structures, related
literature, and embeddings all within the same
environment, ensuring that they have the
LLMs for hypothesis generation and insights that might not be immediately
decision support apparent, such as identifying unexpected
correlations between different molecules or
Large language models (LLMs) trained on predicting potential side effects based on
extensive datasets of chemical and biological historical data.
information can further enhance the drug
Outcome and benefits
discovery process by generating hypotheses,
proposing new molecules, and optimizing ● Time and cost savings: The integration
reactions. Given a specific query or a set of of AI with MongoDB Atlas significantly
parameters, LLMs can generate novel SMILES reduces the time required for drug
strings for potential drug candidates, predict discovery and reaction optimization,
their properties, and suggest optimal synthesis leading to faster time-to-market and
routes. lower R&D costs.
● Improved accuracy: AI-driven
LLMs also play a crucial role in decision
predictions are more accurate,
support. By analyzing the vast amount of data
reducing the need for extensive manual
stored in MongoDB Atlas, they can provide
experimentation and iteration.
Cutting-edge recommendation The implementation of Exafluence’s solution
engine leverages advanced is expected to reduce the time-to-market
for specialty life sciences products by 40%.
language models and vector search
The enhanced system enables faster and
technology. more accurate molecule discovery,
streamlining the production of specialty
This AI-driven system integrates MongoDB’s chemicals across various industries.
modern data platform to store and manage Additionally, the planned introduction of a
complex molecular data, allowing scientists mobile app will use real-time data to further
to perform natural language queries for improve the accuracy of molecular
rapid identification of promising molecular predictions, enhancing research decisions
structures. The platform also unifies data and overall operational efficiency.
from various sources, including time series Learn more
data from PLC devices, enabling faster and
more refined molecule discovery processes.

Figure 50: Screenshot from Exafluence AI-powered platform showcasing the automated
synthesis planning of DEET using integrated web search and reaction planning tools
Other Notable Use Cases

Behavioral health monitoring Personalized patient history

and sentiment analysis visualization
Generative AI integrated into wearables or Generative AI customizes and summarizes
apps can monitor conditions like depression patient history views based on the specific
and anxiety by analyzing speech, activity needs of each healthcare professional. By
levels, physiological signals, and sentiment. It highlighting the most relevant records, results,
provides early alerts and suggests and treatments, and providing concise
interventions, enabling proactive care and summaries, AI streamlines access to critical
improving mental health management. information.

Automating prior authorization, Personalized treatment

coding, and billing pathways in clinical trials
Generative AI automates prior authorization, Generative AI analyzes patient data to
coding, and billing by assisting with form create personalized treatment plans for
completion, accurate record coding, and clinical trials, predicting responses and
reviewing insurance policies. This reduces optimizing trial designs. This improves
errors, speeds up approvals and efficiency, accelerates drug development,
reimbursements, and improves efficiency for and enhances patient outcomes.
both providers and payers.

Contact
Information
Francesc Mateu Amengual
Healthcare Industry
Solutions Principal
[email protected]
AI Leaders
in Action
Hear from AI pioneers sharing their expert
insights on the future of AI
 Ben Ellencweig,
McKinsey senior partner
and global alliance leader
for QuantumBlack, AI by
McKinsey

Can you share a bit about your career

Ben Ellencweig is a McKinsey senior partner journey and what led you to become a
who leads alliances, acquisitions, and McKinsey Senior Partner and leader in
partnerships globally for QuantumBlack, AI by QuantumBlack, AI by McKinsey?
McKinsey. Ben advises clients on harnessing
the power of AI to deliver growth and My work at McKinsey goes back nearly 15 years
bottom-line impact across sectors. at this point, starting out as an associate after
getting my MBA in 2009. During that time, I’ve
worked on a broad range of technical challenges,
applying new technologies to industries ranging
Quantumblack, McKinsey’s AI from automotive to software to media. More

arm, supports transformation recently, I’ve been focused primarily on AI, and
am currently the global leader of alliances and
initiatives using the power of
acquisitions for QuantumBlack, AI by McKinsey.
technology, technical expertise,
I also led our acquisition of Iguazio, a leader in
and industry experts.
AI and machine learning technology, in 2023.
QuantumBlack Labs is the R&D
and software development arm Before McKinsey, I spent nearly a decade in roles
of QuantumBlack. leading the deployment of new technologies in
government and military settings. The constant
theme throughout my career has been
implementing cutting-edge technology in
innovative ways that transform large
organizations and create significant impact at
scale. This is why it makes sense that I've
become focused on AI—there’s simply no other
technology in our lifetime with the same kind of
potential to improve society and our lives.

US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
QuantumBlack has been instrumental in
leveraging AI across various sectors. Can
you highlight some of the most impactful
AI projects or initiatives led by
QuantumBlack?

QuantumBlack has led some truly incredible

initiatives that have transformed major players
in a variety of sectors.

● We helped a major pharmaceutical

company develop an AI-powered drug
discovery platform, which accelerated the
identification of promising drug candidates
by 60%.
● In manufacturing, QuantumBlack worked
From your experience advising Fortune 500
with a global automaker to implement an
executives and boards, what are the most
AI system for predictive maintenance that
common misconceptions about AI in the
reduced unplanned downtime by 30%.
business world?
● We built an AI-driven risk assessment
model for a financial services client that
There are several! The most common
improved fraud detection rates by 40%,
misconception is one that shows up with nearly
while reducing false positives.
any emerging, transformative technology: the
● In healthcare, our AI solution for optimizing
belief that it can be used as a plug-and-play
hospital operations led to a 15% reduction
solution, rather than one that requires strategic
in patient wait times and 20% improvement
integration. Executives often look for immediate
in resource utilization.
ROI from AI, without accounting for the
There are also some high-profile projects that necessary investments in data infrastructure,
we can mention more openly, like developing talent, and change management. I also see a
an AI-enabled “dashboard” that led to a world tendency to underestimate the importance of
speed record for a Formula E electric car high-quality, well-governed data as the
racing team. foundation for effective AI.

I’m especially proud of our ongoing work with It’s all too easy to overestimate AI's capabilities
One Ocean, a foundation that leads research in the short-term, expecting human-level
efforts to support the development of an reasoning across all domains, even if it’s working
eco-friendly “blue economy.” Identifying which with low-quality or poorly structured data. In
companies are truly having a positive terms of gaining long-term value from AI, I’d say
environmental impact has always been the biggest mistake is executives focusing solely
difficult, so we helped One Ocean develop an on productivity through automation and
AI-powered analytics platform that cost-cutting rather than AI's potential for
streamlines the process dramatically. business model innovation and value creation.
Given the varying rates of generative AI Data is a critical component of successful
adoption across industries, what specific AI implementations. What are the unique
industries do you see as leading the charge data challenges and privacy
in terms of deriving maximum value from considerations for different industries, and
the technology? What are the key factors how can organizations effectively address
driving this disparity? them?

Right now, technology, financial services, and One that comes immediately to mind is
healthcare are at the forefront when it comes healthcare: There’s tremendous value to be
to harnessing the power of gen AI. Tech gained, but you’re also dealing with incredibly
companies are leveraging it for software sensitive patient data that requires robust
development, content creation, and customer anonymization and consent management
service. Financial services firms are using it for practices. Both healthcare and financial
personalized financial advice, risk assessment, services face strict regulations on data usage,
and fraud detection. Healthcare is seeing and finance in particular has to worry about
significant applications in drug discovery, cross-border data flows. Retail and
medical imaging analysis, and clinical decision e-commerce face similar but less extreme
support. issues, as the more they use data to
personalize experiences, the more privacy
Although these three sectors are very different, concerns they bump up against. Regulations
they have a few things in common that have like GDPR have already made these concerns
contributed to their early adoption. All three broadly understood, but there’s still plenty
are very data-driven with well-developed being worked out. Any sector that deals with
technological infrastructure, meaning they classified or proprietary information, including
have access to the well-structured, high-quality manufacturing, government, and defense,
data needed for meaningful AI action, and must put serious effort into ensuring that data
may not need to make big investments in is protected.
hardware. Software and finance are also
sectors where AI-savvy talent is already Every sector faces issues of data quality and
abundant. They have clear use cases integration to some degree, and very large
(mentioned above) that are closely aligned industries with a wide range of customers (like
with core business processes. Last but not healthcare,) often struggle trying to manage a
least, these are competitive sectors that are variety of datasets with different formats and
used to making investments in order to quality levels. Strong data governance
maintain an edge, and stay ahead of the pack. frameworks can go a long way to addressing
these issues, and data quality and integration
tools are getting better all the time. There’s
also been rapid improvement in privacy-
preserving technologies like federated learning
and differential privacy. Ultimately, good data
management is a strategic and even cultural
issue, and one that’s essential for AI success.
How do you envision generative AI QuantumBlack emphasizes the integration
reshaping specific industries in the next of hybrid intelligence, combining human
five to 10 years? What are the key trends expertise with AI capabilities. Can you
and disruptive innovations to watch for? elaborate on how this approach has been
implemented in specific projects or
Predicting the future is always a challenging sectors? What benefits have organizations
task, but if I had to guess I think some of the experienced from leveraging this blend of
most visible changes will be in healthcare, with human and artificial intelligence, and how
AI-assisted drug discovery, personalized does it enhance decision-making and
treatment, and AI-powered diagnostic tools. operational efficiency?
You can expect more effective, more targeted
medical treatments, and potentially a That’s true, hybrid intelligence is something
reduction or at least a leveling of costs, which QuantumBlack has championed since its
all have exciting implications for our quality of beginning and successfully implemented
lives. Several other industries will also start across a wide range of projects and sectors. To
offering hyper-personalized products and give one example, we integrated AI-powered
services combining text, image and video, predictive maintenance models with human
from financial services to education to expert knowledge in a large oil and gas project
shopping, in ways that we have never to optimize equipment maintenance schedules,
experienced before. which reduced downtime by 25% and brought
significant cost savings. For a financial
I'm quite excited to see how the intersection of services client, we developed an AI system
AI and blockchain impacts legal services, and that assists human analysts in detecting
how IoT sensors can be leveraged through AI complex fraud patterns, improving detection
analysis to revolutionize manufacturing, rates by 35% while reducing false positives. In
realizing unprecedented levels of efficiency healthcare, we helped one provider with a
and predictive maintenance. hybrid system for patient triage: AI provides
initial assessments that are then refined by
human clinicians, leading to patient routing
that’s 20% faster, and more accurate as well.

In each of these cases, the human operators

welcomed this kind of AI support, seeing it as
an extension of the technological tools they
already used and a seamless way of
augmenting their expertise.
What key skills and qualities do you
believe are essential for future leaders in
AI?

Given how rapidly the AI field is evolving, and

will continue to evolve, leaders will need to be
exceptionally flexible and adaptable, and able
to learn continuously. This is in addition to
other qualifications that might be more
obvious, like an understanding of data science
and machine learning, and an ability to think
strategically in order to identify high-impact AI
use cases and align them with business
objectives.

Another aspect of AI that demands a unique

skillset is its breadth. Because it is a
general-purpose technology with the ability to
touch almost any kind of human pursuit, AI
needs leaders with a global and
multidisciplinary mindset who are able to
translate technical concepts for stakeholders in
non-technical fields. Finally, it should be clear
that AI brings tremendous ethical
considerations with it, and that anyone seeking
to lead here needs to not just respect ethics
but do the work of understanding how they
play out in real people’s lives.
 Asaf Somekh,
Founder & CEO
at Iguazio (acquired by
McKinsey)

Asaf Somekh is the Founder & CEO at Iguazio Iguazio offers an AI platform
(acquired by McKinsey) that enables enterprises to
develop, deploy, and manage
Can you share your personal journey in the ML and gen AI applications at
world of AI that led you to found Iguazio?
scale. It serves as a unified
What were the specific challenges you
place for data ingestion,
encountered in the industry that inspired
storage, processing, model
you to create a platform to address them?
training, and deployment.
In late 2014, after the successful acquisition of
Voltaire (which went public on NASDAQ in 2007
and was later on acquired by Mellanox, now a
part of NVIDIA) I climbed Mount Kilimanjaro In early 2023, after eight years of building and
with one of my co-founders. Together, we scaling the company, Iguazio was acquired by
decided to create Iguazio to address the McKinsey. Today, Iguazio is a part of
challenge of operationalizing and de-risking QuantumBlack, McKinsey’s AI arm. Together
AI. We saw firsthand the challenges that with McKinsey, we help global organizations,
enterprises were struggling with when trying to including Fortune 500 companies, with special
implement machine learning in the interest from with the financial industry,
organization. Many models were either unable implement and scale gen AI responsibly.
to reach production or, worse, introduced
errors and risks that had a negative impact on Could you explain how Iguazio’s platform
the business. These weren’t just technical simplifies the development, deployment,
setbacks but rather financial losses, and management of gen AI applications?
inefficiencies, and missed opportunities for
innovation. At Iguazio, our mission is to bring AI to life,
making AI implementation not just possible,
ChatGPT galvanized the adoption of AI in but scalable, effective and responsible in live
2022, making the challenges of bringing AI business environments across industries.
from development to production and impact
even bigger.
US 866-237-8815 | INTL +1-650-440-4474 | [email protected] | © 2023 MongoDB, Inc. All rights reserved.
We focus on two critical aspects that set our
platform apart: operationalization and
de-risking.

We've built our platform to automate and

streamline the entire AI lifecycle, from data
ingestion and model development to
deployment and continuous management.

This means you can take a generative AI

application from a proof of concept to
full-scale production quickly and reliably. Our
platform handles the heavy lifting by
orchestrating four robust processes covering
everything from development to production:
data management, LLM training and
customization, application deployment, and
monitoring and governance. We also ensure An AI factory operates on the principle of
that your AI applications are highly scalable, continuous, automated and scalable
with optimized GPU provisioning, and flexibly production—much like a traditional
deployed in any environment—on-premises, manufacturing factory but for AI applications.
multi-cloud, or hybrid. The AI factory approach ensures that gen AI
and ML models are always up-to-date,
While generative AI offers incredible
performing optimally, do not pose risks, and
opportunities, it also comes with significant
are aligned with business objectives.
risks, such as AI hallucinations, IP
infringement, biased outputs, data privacy These models are governed with the data sets
concerns, and more. Iguazio’s platform allows that were used to train each version, together
enterprises to easily add guardrails to mitigate with the application source code. This is done
these risks across the entire AI pipeline. by applying clients’ CI/CD frameworks but
with these extra dimensions. This is critical for
Putting this all together is essentially the “AI
addressing the AI regulations that are being
Factory” concept that we’ve been evangelizing
introduced globally, starting with the EU. The
even before the gen AI hype.
AI Factory also addresses the growing
complexity in AI by integrating various
With the growing complexity of AI models
processes—such as data ingestion, model
and the need for continuous improvement,
development, validation, deployment, and
why do you believe the concept of an AI
monitoring—into a unified pipeline that
factory is essential for scaling and
reduces friction and inconsistencies. For
maintaining AI systems effectively? How
organizations that want to maintain a
does Iguazio's platform embody this idea
competitive edge, this enables them to
to streamline the end-to-end AI lifecycle?
combine gen AI with traditional AI, rapidly
iterate on AI models, and deploy new
capabilities in response to changing market
conditions or emerging data.
The Iguazio AI platform is built to streamline All your data needs are covered by MongoDB
the end-to-end gen AI and ML lifecycle. from vector search to a document database to
SQL queries, and Iguazio’s AI Factory is
We provide an integrated environment that layered on top.
accelerates the deployment of foundation
models, transformers, and application logic in This partnership and pre-made recipes provide
production, while maximizing GPU enterprises with the capabilities to
performance. Comprehensive testing and operationalize AI with minimal engineering,
built-in monitoring and management from prototype to production to monitoring.
capabilities ensure the quality and reliability of They can also scale their data while ensuring
user-facing applications. We built and high performance, reliability, and accuracy.
maintain open-source AI orchestration Customers can rest assured with enterprise
framework MLRun, which is at the heart of our security and compliance, including encryption,
platform, for the community to further support access controls, and compliance monitoring.
these workflows. So enterprises can
collaborate efficiently, streamline operations, In addition, we support hybrid environments
and accelerate their time-to-market. and provide unification of all data
management needs, like logging, auditing, and
Iguazio and MongoDB have teamed up to more to streamline the process and ensure
create a powerful solution for gen AI consistency and faster performance. Finally,
applications. Could you explain the we can accommodate diverse use cases, like
synergy between the two platforms and the real-time co-pilots for different professions,
benefits it offers to enterprises? virtual agents, call center analysis applications,
chatbots, and more. With McKinsey’s expertise
MongoDB is known for its flexibility, scalability, in strategy and consulting, we can also guide
and ability to handle vast amounts of various organizations strategically throughout their
types of data—key requirements for building AI-based transformations.
and scaling generative AI applications. On the
other hand, Iguazio brings to the table a How does Iguazio’s platform support the
sophisticated platform that automates and seamless integration of LLMs and
optimizes the AI lifecycle, from data traditional ML models, and what are the
preparation and model training to deployment key considerations in doing so?
and continuous monitoring and de-risking.
This is a super important question. We see
What I also love about this partnership is that
organizations build separate environments and
both MongoDB and Iguazio have always been
teams for gen AI and traditional AI. Also, we
focused on scalable production environments.
see too many organizations trying to use LLMs
The integrated MongoDB-Iguazio solution for AI applications. These are the wrong
allows customers to streamline data approaches. Gen AI with its LLMs will
processing and the AI application life cycle. sometimes underperform compared with a
Together, they ensure gen AI applications standard machine learning model or even
reach production while eliminating risks, simple heuristics. Therefore, the right approach
improving performance and enhancing is to build pipelines that can combine the
governance. adequate technique in each step of the
application.
Iguazio’s platform is uniquely designed to For example, in customer engagement,
support various AI needs, like the integrated capabilities can include call center coaching,
use of both LLMs and traditional ML models, all agents, virtual co-pilots assisting human
while addressing the complexities of managing agents, segment identification, and
and deploying both types of models within a coordinated outreach campaigns.
unified framework.
For finance, future capabilities include
Example use cases include integrating a advanced market intelligence and synthesis,
customer-facing chatbot with a classifier model and M&A target identification and strategy.
to check information about customer queries,
generating synthetic data for ML model But, as I mentioned before, AI models also
training, or using sentiment analysis for feature introduce risks, which will need to be
engineering. But there are many more. addressed. For example, producing unfair and
biased outputs, IP breaching, PII and data
Using LLMs and traditional ML involves the privacy breaching, poor LLM accuracy,
same MLOps (machine learning operations) hallucinations, toxic responses, and
practices. This includes ensuring you address compliance violations. On top of that gen AI
automation and coordination of workflows like can become a very expensive exercise if the
data preprocessing, model training and infrastructure is not used correctly. We’re
deployment, ensuring flexible access, already seeing enterprises building gen AI
distribution of GPU resources, and apps where operating costs are higher than
computational power when needed, real-time the value they bring.
capabilities, monitoring for model performance,
and establishing guardrails. For these reasons, it’s important that
enterprises build their architecture in a
Given the rapid advancements in gen AI, forward-thinking manner, maintaining
where do you see the most significant flexibility and the ability to swap components
opportunities and risks for enterprises over in pipelines, involving open-source
the next few years? technologies, introducing guardrails, and
supporting various LLM customization
Over the next few years, I see several key areas capabilities, like RAG and fine-tuning, to allow
where businesses can either thrive or face their architecture to change and evolve
significant challenges, depending on how they together with the rapidly evolving landscape.
navigate this new era. For example, McKinsey &
Company has identified that gen Al has the Can you provide specific examples of how
potential to deliver an additional $200-340 Iguazio's platform has helped enterprises
billion in value for the banking industry alone. in a specific industry, such as finance,
The top use cases they see adding value across healthcare, or retail to achieve tangible
industries are using gen AI as a virtual expert, business outcomes?
for content generation, for customer
Iguazio is trusted by enterprises across
engagement, and accelerating coding.
multiple industries: finance,
But it’s not just about automation and telecommunications, manufacturing, mobility
acceleration, the entire value chain can be and more.
transformed.
McKinsey acquired Iguazio in 2023 after a For organizations just starting with gen AI,
thorough market scan looking at hundreds of what advice would you give them to ensure
players in the field. Seeing that 90% of AI they maximize their ROI while mitigating
projects fail to deliver impact, and following risks?
extensive internal research into the solutions
available in the market, they determined For organizations just starting with generative
Iguazio is best placed to scale AI and gen AI AI, there are several key steps to take to
across the organization, put thousands of gen ensure maximum ROI and minimum risk. The
AI apps into production, and, for the same first step is all business.
investment, implement more projects, models,
and impact for the enterprise. Begin by identifying and prioritizing the
specific business problems you want to solve
One of our client is a fast growing European with generative AI. Whether it’s improving
digital bank with 3.5 million customers and customer service or analyzing activity.
120% YoY growth. Using Iguazio, they built a
gen AI call center analysis app to improve call Now, it’s time to choose the right technology
center operations, simplify agent training, to support your business needs, such as a
improve customer experience, and reduce cost. platform that can support your business goals
and help you leverage AI to derive tangible
The gen AI application, deployed on-premises, business value. Depending on your needs and
summarized customer calls, analyzed current infrastructure, it will often make sense
sentiment and topics, and removed personally to select a platform that supports the entire AI
identifiable information (PII). The data was fed lifecycle—from data management to model
to downstream applications—like live agent deployment and monitoring. A platform like
support, customer profiles, auto-generated Iguazio’s, which serves as a complete gen AI
content, tailored recommendations, and factory, allows for seamless and automated
customized offers. scaling and operationalization of AI models,
accelerating time to market, and ensuring that
your AI initiatives are both efficient and
cost-effective.

Finally, establish clear governance frameworks

to oversee gen AI model development and
deployment. This includes setting up guardrails
for regular model validation, bias detection,
and performance monitoring.

This is a cycle, and you can and should always

use feedback and monitor results, as well as
update your models and applications as
business needs change. These steps will ensure
your AI projects are continuously bringing
value to your business.
AI and Application
Modernization
Programs
AI modernizes apps faster for a better
user experience (UX)

122
MongoDB Atlas for
Industries: Innovation
Workshops

An innovation workshop is a carefully

curated, virtual or in-person event where
MongoDB industry experts and decision
makers from an account come together
to discuss the “art of the possible”
relating to MongoDB-based solutions
Innovation Workshops
Accelerate mission-critical applications, drive
innovation, and transform AI challenges into
opportunities with unparalleled speed,
reliability, and scalability

An Innovation Workshop is a carefully curated virtual or in-person event during which

MongoDB industry experts and decision makers from customers and prospects come
together to discuss the “art of the possible” relating to MongoDB-based solutions. Explore
bespoke, modern app development and tailored solutions backed by our extensive industry
knowledge and best practices. Discover exclusive workshops with MongoDB's top experts.

Accelerate your
modernization
journey Industry-specific solutions The art of the possible
Gain insights into how MongoDB and Explore the art of the possible with
adjacent technologies can provide MongoDB-powered solutions. Learn
end-to-end solutions for from real-world use cases that
Learn more
industry-specific challenges. demonstrate how MongoDB can
Understand how similar issues have transform data management and
been addressed successfully by other modern application development
leading companies in your field. within your industry.

Collaborative discussions Execution excellence

Engage in discussions with MongoDB Understand MongoDB’s project
experts and technology partners. execution capabilities, including
Learn how others in your industry successful collaborations with
have leveraged MongoDB to solve technology partners. Learn how we
complex problems and drive can help you achieve your business
business growth. goals with efficiency and expertise.
MongoDB AI
Applications Program

The MongoDB AI Applications

Program (MAAP) helps organizations
rapidly build and deploy modern
applications enriched with AI
technology at enterprise scale
MongoDB AI
Applications Program
Rapidly build and deploy modern
applications enriched with generative
AI technology at enterprise scale

MAAP is a comprehensive program designed to accelerate gen AI application development

for companies at every stage of their AI journey. It combines the technologies and services
to progress customers from idea to roadmap to prototype to production. Molded to fit each
organization’s unique goals, timeline, and budget, customers are guided on the shortest
and strongest path to success with a bespoke combination of resources.

Accelerate
your AI
journey MAAP Ecosystem Composable Architectures
Technology and expertise from Pre-designed architectures serve
the world’s leading AI and tech as accelerated frameworks for
organizations, across the entire fully customizable and secure
gen AI stack, for a cohesive, applications to accommodate
fully supported application ever-evolving gen AI use cases.
development experience.

Expert Services Education & Enablement

MAAP offers strategic guidance Comprehensive resources enable
on roadmaps and skillsets, developers to quickly—and
assists with data integration into repeatedly—build modern AI
advanced AI technologies, and applications faster while giving
helps organizations develop organizations the tools to expand
production-ready applications. their in-house AI expertise.
The MAAP to AI success
Access to trusted partners and a robust support To directly address this need, MongoDB created
system are critical for organizations that want to MAAP, which offers customers a clear path to
create a competitive advantage with AI in today's developing and deploying AI-enriched applications
fast-paced innovation economy. Leaders are through reference architectures—an end-to-end
excited about gen AI, but are also unsure how technology stack that includes integrations with
to move from concept to production and want to leading providers, professional services, and a
control costs. unified support system. The cornerstone of MAAP is
MongoDB technology. Applications are underpinned
Many early adopters have struggled with the by MongoDB, which securely unifies real-time,
inefficiencies of legacy technologies that cannot operational, unstructured, and AI-related data
scale or manage the real-time, multi-modal data without the need for bolt-on solutions.
required to power AI-enriched applications. This is
compounded by a lack of in-house AI development What’s more is that customers can use any LLM
expertise and the perceived risk of integrating provider and run anywhere (on all major cloud
disparate components without coordinated support. providers, on-premises, and at the edge). MongoDB
Due to these challenges, businesses aren’t able offers seamless integrations with the entire
to put AI-powered applications into production ecosystem of AI frameworks and solutions.
to deliver innovative, intelligent solutions to
their customers. Perhaps most importantly, thousands of
customers already rely on MongoDB to power
Capturing the AI opportunity and cohesively their mission-critical apps, and we have years
bringing the cutting-edge innovation happening of experience helping customers unlock the power
in the AI space directly to customers requires of data. Ultimately, MAAP aims to help customers
the right technology, investment, and collaboration confidently innovate so that they can accelerate
from industry-leading providers across each layer time-to-market, minimize risks, and realize the value
of the gen AI application stack. of AI investments sooner.

The AI Stack

Learn more
 While the MAAP ecosystem is just getting
MAAP started, it already includes industry leaders
from every part of the AI stack, including
Ecosystem Anthropic, Anyscale, Arcee AI, AWS, Cohere,
Credal, Fireworks AI, Google Cloud,
LangChain, LlamaIndex, Microsoft Azure,
Nomic, and Together AI.

The result is a group of organizations that

will enable customers to build
differentiated, production-ready AI
applications—with simplified technology
selection, reduced integration complexity,
and optimized compatibility—while aiming
to deliver substantial return on investment.

The AI Stack
The cornerstone of MAAP is MongoDB technology. MongoDB sits at the heart of the AI
application stack, the data layer, removing the friction of integrating, securing, and maintaining
the essential data components required to build AI applications. MongoDB underpins these
applications, securely unifying real-time, operational, unstructured, and AI-related data without
the need for bolt-on solutions. MongoDB’s open and integrated architecture provides easy access
to the MAAP ecosystem and enables the extension and customization of applications to tackle
any use case.
The MongoDB AI Applications
Program and its ecosystem of
companies offers customers the
right expertise and solutions for their
use cases. It removes integration
risk, making it easy for businesses to
use the industry-leading
technologies of their choice to
unlock the value of AI with their
data. Learn more about each
member of the MAAP ecosystem in
the MongoDB Partner Ecosystem
Catalog.
 The MAAP framework provides access to a
Composable set of downloadable, pre-designed,
pre-integrated RAG architectures with
Architectures pre-configured UIs that serve as
accelerated frameworks for building AI
applications. Architectures are fully
customizable and extendable to
accommodate ever-evolving generative AI
use cases, like retrieval-augmented
generation (RAG) or advanced AI
capabilities like Agentic AI and advanced
RAG technique integrations. With
MongoDB’s open and integrated platform
at its core, innovation with MAAP’s
composable architectures is unlimited.

Easily bring the power of leading AI platforms directly to your

generative AI applications

Figure 51: The MAAP framework is a set of libraries that you can use to build your RAG application
using MongoDB and Atlas Vector Search. Above is the reference architecture diagram of the
framework with various components
 MAAP expert services, combining the strengths
of MongoDB Professional Services and
industry-leading gen AI consultancies, will enable
Expert Services customers to rapidly innovate with AI. MAAP
offers strategic guidance on roadmaps and
skillsets, assists with data integration into
advanced AI technologies, and can even develop
production-ready applications.
MAAP goes beyond development, empowering
teams with best practices for securely
integrating your data into scalable gen AI
solutions, ensuring businesses are equipped to
tackle future AI initiatives.

Professional Services can accelerate

development of AI applications and get hands
on, taking solutions into production. Learn more
about AI Accelerators and Application
Development Services.

MAAP boutique and global system integrator

partners include gravity9, PeerIslands,
Pureinsights, and Accenture.

A major car maker in EMEA teams up with MongoDB Professional

Services and Pureinsights to revolutionize noise diagnosis.
The Noise Diagnostics Application prototype—based on Atlas Vector Search and Google’s Vertex AI
engine—uses car noise to assist mechanics perform root-cause diagnostics faster and more
accurately, reducing dealership diagnosis time and vehicle immobilization. It also helps identify
potential design issues and reassures clients about a car's condition, reducing unnecessary visits to
the dealership. This initiative not only cuts warranty costs but also enhances customer satisfaction
and brand image for this leader in the automotive industry.

Built an Atlas Vector Search 83%

solution to search noise Of target noise reports appear
reports in top 10 results

Evaluated cross lingual 73%

searches in 11 languages
Mean accuracy
Evaluated 4 embedding
models to ensure best 2
performance with 32 Weeks evaluation time
automated tests
 MAAP customers have access to a variety

Education & of learning materials, including a dedicated

MAAP GitHub library featuring integration

Enablement code, demos, and a gen AI application

prototype. These comprehensive resources
enable developers to build intelligent,
personalized applications faster, while
giving organizations the tools to expand
their in-house AI expertise. With MAAP,
customers have access to integration and
development best practices that they can
use for future gen AI projects.

In addition to digital resources, our

MongoDB .local series offers in-person
opportunities to learn how to accelerate the
delivery of production scale and safe
AI-enriched apps.

Learn from experts and connect with community

MAAP Framework
The MAAP Framework is a set of libraries that you
can use to build your RAG application using
MongoDB, Atlas Vector Search, and associated
MAAP ecosystem partners.

MongoDB University
Learn to get the most from MongoDB products
with our labs, courses and live training. Create a
foundation with the Introduction to AI and Vector
Search course or advance your learning with
instructor-led training in AI and Vector Search
Basics.

Solutions Library
Drawing from experience with over 49,000
customers, the Solutions Library is curated with
tailored solutions to help developers kick-start
their projects. Get inspired by real gen AI solutions
spanning diverse industries.

AI Resources Hub
View our library of articles, analyst reports, and
case studies designed to help you build
AI-powered applications.
AI Partnerships
How partners are leveraging MongoDB to
build AI solutions
AI with MongoDB and
hyperscaler cloud services

MongoDB Atlas on AWS, Microsoft

Azure, and Google Cloud

134
Amazon Web Services
(AWS)
Unlock the full potential of generative AI-powered applications with MongoDB Atlas and AWS.
MongoDB Atlas on AWS allows you to build intelligent applications that are highly available,
performant at global scale, and compliant with the most demanding security and privacy standards.

Build fast Build simply

Use MongoDB Atlas Vector Search and Amazon Enjoy fully managed RAG and a unified
Bedrock to build and privately customize LLMs interface and API for all data and application
with real-time operational data. services. Eliminate the need for a bolt-on vector
database and bespoke data pipelines.

Build efficiently Build safely

Get up to 60% faster query times and optimize Integrations with AWS PrivateLink and other
costs when you isolate and scale gen AI AWS services allow you to securely use
workloads independent of the core operational proprietary data with gen AI across the
database using MongoDB Atlas Search Nodes. organization.

Fully Managed RAG in Minutes Gen AI at Novo Nordisk

MongoDB Atlas is a Knowledge Base for “Using Amazon Bedrock and MongoDB
Amazon Bedrock, making it even easier to Atlas… we are the first in the industry to
build generative AI applications backed by generate complete Clinical Study Reports
enterprise data.
in minutes rather than weeks.”

With the click of a button, Amazon Bedrock

“We are doing it at scale, and with just a
integrates MongoDB Atlas as a vector
fraction of the resources we needed in the
database into its fully managed,
end-to-end retrieval-augmented past. It is a game changer for healthcare
generation (RAG) workflow. around the world.”
Learn More
Louise Lind Skov
Head of Content Digitisation at Novo Nordisk
Microsoft Azure

MongoDB Atlas on Microsoft Azure empowers enterprises to build intelligent applications that drive
real-world results derived from your data. Integrations with key Azure services, like Microsoft Fabric,
offer a seamless, scalable, and secure platform to unify your data and launch experiences that
harness the best of AI/ML and gen AI.

Architectural simplicity Dedicated scalability for Gen AI

Combine operational data, vector data, and MongoDB Atlas’s dedicated search and vector
metadata in a single database, then use that search nodes allow you to dynamically scale AI
data with integrated Microsoft Azure services workloads on Microsoft Azure’s global
like Microsoft Fabric, Semantic Kernel, and infrastructure.
more.

Connected platform Security and compliance

There are several ways to quickly and securely Enterprise-grade security features and
connect MongoDB Atlas to Microsoft Azure’s extensive compliance certifications from Azure
suite of gen AI services, including via Fabric and MongoDB.
and Azure AI Studio.

MongoDB Atlas and Microsoft Fabric An Intelligent Ecosystem

A seamless integration with Microsoft Fabric Enterprises need to take advantage of gen AI,
enables you to run large scale AI/ML, analytics, AI/ML, and analytics to get the most from their
and BI reports across your unified data estate data.
on MongoDB Atlas.
MongoDB Atlas, Microsoft Fabric, and Azure AI
Reimagine how teams work with data by Studio operate as an ecosystem, driving
bringing everyone together on a single, actionable intelligence on historical data and
AI-powered platform designed to simplify and real-time intelligence to power AI/ML and gen
accelerate intelligent application development. AI use cases.
Google Cloud

MongoDB Atlas and Google Cloud bring together the two halves of the gen AI equation—data and
AI—to help organizations pursue the full potential of gen AI for their applications and software.
The integration between Atlas and Vertex AI, Google Cloud’s unified ML and AI platform, empowers
organizations to unlock the value of their data for AI, gen AI, and ML.

Creating and storing embeddings Building RAG workflows

Via the MongoDB API, Vertex AI vectorizes Using the Vertex AI platform, developers can
your data, generates embeddings, and passes perform CRUD (create, read, update, delete) on
them back to your MongoDB database for MongoDB Atlas.
storage and retrieval, keeping your data secure
and apart from the LLM dataset, yet readily
accessible for model augmentation.

Streamline the gen AI Stack Develop in natural language

MongoDB Atlas seamlessly integrates with Your teams can query MongoDB Atlas in
Google cloud infrastructure, simplifying the natural language within Vertex AI with Google
connection between your data source and gen Gemini. Automatically turn human language
AI models. into MongoDB-specific query syntax.

MongoDB Atlas and Vertex AI Serving Up AI for Delivery Hero

By harnessing the capabilities of MongoDB “With Atlas Vector Search we can

Atlas and Google Vertex AI, you can confidently
compose sophisticated queries that
deploy generative AI applications and
quickly filter across product data,
overcome the barriers that hold many of your
competitors back. customer preferences, and vector
embeddings to precisely identify
Speed application development, stay at the hyper-relevant product recommendations
forefront of AI innovation, and deliver
in real time.” “We needed to move to an
unparalleled user experiences—all while scaling
up-to-second real-time recommendations
effortlessly, reducing AI hallucinations, and
keeping your data private. system, and that is what MongoDB Atlas
Vector Search enabled us to do.”
Learn More
Mundher Al-Shabi
Senior Data Scientist, Delivery Hero
AI with MongoDB and
System Integrators

Leverage the power of MongoDB

alongside the expertise of Capgemini,
Accenture, Pureinsights, gravity9, and
Peerislands to achieve faster
time-to-market, improved data
management, and enhanced
scalability for AI initiatives
Capgemini
Capgemini and MongoDB have joined forces to provide gen AI-powered solutions that accelerate
businesses' AI journeys. Together, they've developed over 10 industry-specific accelerators to help
organizations harness the power of AI more efficiently.

Genyoda Health druid

Gen AI-powered digital assistant Revolutionizing the healthcare industry with
revolutionizing the way insurance companies AI-powered insights, enhancing the patient
interact with their customers. journey through personalized and
compassionate care.

Customer analytic record Virtual store assistant

Enables successful, enterprise-wide Transforming retail journeys with intelligent
hyper-personalization for banking, wealth virtual assistants for tailored shopping
management and fintech customers with experiences across platforms and channels for
humanized AI chatbots to streamline issue higher conversion rate.
resolution and optimize customer interaction.
Card-not-present fraud prevention Trusted Vehicle
Tackle real-time fraud in card-not-present Accelerate time-to-market with a secure and
transactions through a comprehensive system scalable platform of next-generation driver and
utilizing 3D secure protocol, behavioral fleet-management experiences..
biometrics, user behavior analytics, and
“Building data-driven applications in
machine learning algorithms. This event-driven
which AI is embedded is a new frontier
fraud detection and prevention solution is for many organizations, one that
powered by MongoDB. presents new operational and
Database migration-as-a-service application development challenges,
and requires a new way to think about
Database Convert & Compare (DCC) is a
data at the application level. MongoDB
powerful tool developed by the Capgemini to
has some really good foundations to
optimize activities like database migration,
start making that happen, and when
data comparison, validation, and much more in combined with an Operational GenAI
a database migration roadmap. When accelerator like RAISE gives you an
migrating from RDBMS to MongoDB, DCC entirely new collaborative tool chain
achieves 70% automation and 30% manual that is designed for building these next
retrofit on a database level. generation digital applications.”
Learn More
Steve Jones
EVP, Data Driven Business & GenAI,
Capgemini
Accenture
Together, MongoDB and Accenture provide unparalleled expertise to help customers modernize their
environments and adopt a cloud-first approach throughout their organization. As the cloud becomes
more critical to managing data at scale, our partnership helps enterprises unlock data from legacy
data warehouses and lakes to build new applications faster.

Lift and shift Legacy modernization

Transition your organization from a Reimagine legacy monolithic applications to
self-managed cloud to a multi-cloud continuum, harness the true potential of the cloud and
meeting modern storage needs while still accelerating your shift to modern, scalable
preserving traditional enterprise capabilities. architectures.

Data modernizer tool Smart data mover

Experience faster MongoDB adoption with Leverage a comprehensive toolkit designed to
accelerated data modeling while enabling seamlessly migrate data from on-premise data
seamless migration of applications to the cloud. warehouses to cloud data repositories.

Creating a culture of experimentation Migration made easy

Thomas Edison's 10,000 failures led to 10,000 “Accenture sees growing demand from
successful ways that didn't work. Organizations
companies for solutions that can easily
need this spirit of experimentation to innovate.
migrate applications and data migrations
Mark Porter, MongoDB CTO, and Michael Ljung,
Global Software Engineering Lead and Chief to the cloud. Our Smart Data Mover
Software Engineer for Accenture, discuss how solution accelerates this process for
to encourage a culture of experimentation and moving data into MongoDB and other
normalize failed experiments.
target platforms. We continue to invest in
the Smart Data Mover application and
look forward to expanding our
capabilities using MongoDB as a target
platform.”
Learn More
Shail Jain
Global Managing Director, Data & AI at
Accenture
MAAP Partner
Pureinsights
Pureinsights transforms the way organizations interact with information, leveraging cutting-edge AI
technologies like Generative AI, Vector Search, and NLP to build intuitive, human-centered
applications that go beyond traditional search. From information retrieval to innovative use cases
like audio diagnostics, we help businesses uncover new insights and drive innovation.

Pureinsights, MongoDB's premier services partner for search and AI, has built one of the best
Retrieval Augmented Generation (RAG) architectures for a large European car manufacturer using
MongoDB's tech stack and by leveraging our comprehensive ecosystem for AI. Pureinsights’
expertise in search and AI, along with the Pureinsights Discovery platform, have demonstrated the
capability to help customers take their ideas for search and AI from business idea, to working
prototype, to phased production rollout – all while delivering business value.

“Pureinsights exemplifies the innovation and dedication to customer needs that define
our leading partners, earning them this year's Services AI Partner of the Year award.”
Alan Chhabra, Executive
Executive Vice President at MongoDB

Accelerating AI with MongoDB Gen AI Initiatives

As a MongoDB BSI partner and MAAP launch ● GenAI Build Essentials is an initial
partner, Pureinsights is a preferred service project assessment engagement to
provider for MongoDB’s GenAI Build initiatives. scope out a plan to build a Generative
The purpose of these initiatives is to accelerate AI (GenAI) solution with
the planning and prototyping of AI-powered Retrieval-Augmented Generation (RAG)
applications on MongoDB platforms which will capabilities on MongoDB Atlas.
result in faster time-to-value for customers. ● GenAI Build Implementations are a
Pureinsights staff have over 15 years of follow-up to Essentials. This service
experience with commercial or open-source leverages Pureinsights’ and MongoDB’s
data, search and AI platform. technical and development expertise to
build a working prototype of the GenAI
solution implemented on MongoDB
Atlas with retrieval-augmented
generation (RAG) using synthetic data
based on your existing datasets in a
MongoDB environment.

MAAP Partner
gravity9
gravity9 is a leading technology partner, driving businesses beyond outdated legacy systems into the
future with AI-led, cloud-based platforms designed for continuous innovation. Our expertise in
application modernization harnesses advanced cloud technologies, artificial intelligence, and
microservices architecture to streamline operations, elevate user experiences, and foster ongoing
growth. Through our sleek micro-UI platform, we empower organizations to thrive in a rapidly
evolving market, ensuring agility, scalability, and a lasting competitive edge.

gravity9 is proud to be one of only six global strategic MongoDB implementation partners, offering
comprehensive solutions that combine MongoDB's powerful data platform with gravity9’s expertise
in Generative AI and data analytics. gravity9 harnesses the power of MongoDB Atlas to deliver
advanced AI solutions and develop applications based on Large Language Models (LLMs), Retrieval
Augmented Generation (RAG), graph knowledge base and GraphRAG. Our expertise also extends to
Agentic AI solutions for workflow automation, driving efficiency and innovation. By integrating these
cutting-edge techniques and leveraging exclusive access to MongoDB resources and insights, we
help businesses embrace the future of AI and data-driven growth with confidence.

Partnering for Success with MongoDB Digital Transformation Experts

gravity9’s deep partnership with MongoDB gravity9’s highly skilled professionals are
enables them to deliver a flexible, best-practice dedicated to transforming digital applications,
approach to technology modernization. Backed rapidly revitalizing outdated systems with
by rigorous training and certification, gravity9’s tailored, innovative solutions. Their approach
team is fully equipped to design, implement, unlocks new possibilities for businesses across
and optimize MongoDB solutions tailored to various industries by delivering efficiency,
each client's needs. Whether migrating, flexibility, and enhanced functionality.
optimizing, or building new applications, they
ensure seamless integration and rapid
deployment, maximizing the value of each
investment.

MAAP Partner
PeerIslands
Since 2018, PeerIslands has been leading the AI revolution with an exceptional
team and state-of-the-art technology, propelling AI-driven software solutions into the Enterprise and
SMB marketplace. Central to our mission is a handpicked team of top-tier, Top 1% developers,
leveraging advanced AI platforms and our distinctive ‘Human in the Loop’ approach. We go beyond
conventional software development to turbocharge your enterprise applications, reimagine outdated
systems, and harness the power of your data.

We work very closely with MongoDB and complement a great database product with end to end
solutions that customers want. With a talent pool of over 150 MongoDB certified Polyglot engineers
and our active collaboration with Enterprise clients, PeerIslands has been named MongoDB's
Boutique SI of the Year for 3 of the last 4 years. Our software delivery expertise, combined with
MongoDB technology and Solution Architects, has been a recipe for success for clients seeking to
modernize their applications and migrate their data to a modern, scalable, and flexible data
architecture.

PeerIslands’ Polyglot developers enhance MongoDB's capabilities by providing comprehensive

solutions that drive client success. Our team boasts extensive experience delivering critical
applications, demonstrating our commitment to excellence.

Payments Platform Modernization Digital Transformation Experts

PeerIslands is currently working closely with a ● Dramatically speed up the transition

leading Payment Services Provider who from legacy platforms, reducing costs
supports the backbone of top financial and providing a 70% time-to-market
institutions that rely on services such as wire advantage for new products and
transfers, real-time payments, and bill services.
payments. They are collaborating with the client ● With MongoDB at the core, their
on a multi-year modernization journey aimed at platform features a data model that
transforming critical business applications supports their leadership's vision for
using MongoDB Atlas. The combination of AI-enabled services for their customers.
PeerAI and MongoDB Atlas has played a crucial
role in accelerating the client’s time to market
and facilitating the transition from traditional
databases.

MAAP Partner
Unlocking the Power
of AI With SaaS

AI and SaaS are each powerful forces

in tech, but when fused together, their
impact can be even greater than
expected. Let's delve into MongoDB
SaaS AI partners that you can
leverage to save building the solution
yourself
Iguazio (acquired by McKinsey) &
MongoDB: Building & scaling gen AI
apps for enterprises efficiently,
effectively and responsibly.
Iguazio (acquired by McKinsey & company) is a Gen AI Factory & MLOps tech stack that accelerates
the development, deployment and management of ML and Gen AI applications. Trusted by large
Financial Services, Manufacturing, Transportation and Retail clients, including Fortune 500
companies, Iguazio ensures that AI and gen AI applications don’t just remain in the lab, but have
real impact in live business environments.

From building your first Gen AI app, to a full blown Gen AI Factory
By automating and streamlining AI, Iguazio accelerates time-to-market, lowers operating
costs, de-risks, provides guardrails and enhances business impact and profitability. This
enables Iguazio to support enterprise needs, either in a self-serve or managed services
model, with an open and flexible architecture.
Iguazio provides you with the latest capabilities for:
1. Gen AI Ops: Operationalizing AI / Gen AI apps efficiently at scale to create real
business impact.
2. Gen AI Guardrails: De-risking Gen AI to meet compliance, regulations and controls
relevant to your industry while ensuring peak performance.
Iguazio supports data management, training and fine-tuning LLMs, application deployment
and LiveOps that enables monitoring models and data for feedback.

Accelerated and De-risked AI & Gen AI Deployment

● AI / Gen AI Operationalization with ● Scalability and performance: enables

minimal engineering: MongoDB and effortless management of large data
Iguazio offer a unified, scalable data solution volumes and intricate transformations,
from prototype to production. ensuring high reliability and accuracy.

● Hybrid environments: MongoDB and ● Security and compliance: MongoDB and

Iguazio offer flexible deployment options: Iguazio ensure top security and compliance
cloud, on-premises, or hybrid, tailored to for finance and other regulated sectors,
meet MLOps/LLMOps and DataOps needs. safeguarding sensitive data with encryption
and access controls.
● MongoDB and Iguazio unify all data
management needs: (logging, auditing, etc. ● Customers build diverse applications and
in a single solution to ensure consistency, derive actionable insights from their data so
faster performance and significantly less they can drive innovation across use
overhead. cases.
Figure 52: How to build and scale gen AI applications efficiently with MongoDB and Iguazio

MongoDB and Iguazio can be used for Guardrails for Protecting Against LLM
creating a smart customer care agent that Risks
documents call details, provides live
Iguazio eliminates LLM risks with guardrails
contextual recommendations as a co-pilot,
that ensure:
provides live agent support, customizes
offers and recommendations and more. ● Fair and unbiased outputs
● Intellectual property protection
First, the joint architecture processes and
● PII elimination to safeguard user
analyzes raw data (e.g., web pages, PDFs,
privacy
images) inputted by the customer or the
● Improved LLM accuracy and
enterprise.
performance for minimizing AI
Then, the data is processed in a batch hallucinations
pipeline for analyzing customer logs and a ● Filtering of offensive or harmful
stream pipeline for live interactions. content
Finally, results are stored in MongoDB, ● Alignment with legal and regulatory
leveraging its capabilities for managing standards
unstructured data like user age, preferences ● Ethical use of LLMs
and historical transactions, together with
structured data like account balance and
product lists.

© 2024 MongoDB, Inc. All rights reserved.

Build safe, reliable,
enterprise-grade gen AI solutions
with MongoDB and Anthropic

Anthropic is a pioneering AI research and product organization, dedicated to developing reliable,

interpretable, and trusted AI systems. As a public benefit corporation, Anthropic is committed to
ensuring the world's safe transition through transformative AI. Their flagship product, Claude, is a
family of trusted AI models designed for enterprise applications across every industry, which
securely connects to company knowledge to help every team benefit from trusted AI.

Reliability Trustworthiness
Claude powers business-critical use cases, Claude combines best-in-class jailbreak
exhibiting lower hallucination rates and higher resistance and misuse prevention to mitigate
accuracy. brand risk for enterprises.

Safety Flexibility
Claude offers robust security and compliance The Claude model family offers a range of AI
features, meets SOC 2 Type II and HIPAA solutions across the price-performance
standards, and ensures enterprise-grade spectrum, providing enterprises with options for
protection. any use case.

Bring enterprise data to gen AI and use a family of flexible foundation models

Anthropic and MongoDB offer collaborative ● Tailored AI for Your Needs: Deploy and
support to help you navigate the complexities scale generative AI applications tailored to
of building safe, reliable, enterprise-grade your use case and leveraging your data.
generative AI solutions. Together, we provide
● Compliance Built-In: Ensure your
comprehensive solutions for enterprises—so you
generative AI apps are secure and compliant
can launch and scale generative AI applications
out of the box.
safely and reliably, customized with your data.
● RAG Made Easy: Abstract the challenges of
building a Retrieval-Augmented Generation
(RAG) workflow with your data.

MAAP Partner
© 2024 MongoDB, Inc. All rights reserved.
 Figure 53: Cost vs intelligence of Claude models.

The Claude model family offers state-of-the-art performance across a wide range of tasks,
allowing users to select the optimal balance of intelligence, speed, and cost for their specific
applications. Claude’s capabilities include:

● Advanced reasoning: Claude can perform complex cognitive tasks that go beyond
simple pattern recognition or text generation

● Vision analysis: Transcribe and analyze almost any static image, from handwritten notes
and graphs to photographs

● Code generation: Start creating websites in HTML and CSS, turning images into
structured JSON data, or debugging complex code bases

● Multilingual processing: Translate between various languages in real-time, practice

grammar, or create multilingual content

© 2024 MongoDB, Inc. All rights reserved.

Agnostiq & MongoDB:
High-Performance Computing for
All
Agnostiq, founded in 2018, is transforming how AI developers and researchers run compute intensive
applications that utilize GPUs and high-performance computing (HPC) resources. With Covalent, a
workload orchestration and infrastructure management platform, developers code their projects
locally and run on serverless hardware, without needing extensive machine learning operations
expertise. Agnostiq’s commitment to interoperability and resource neutrality ensures accelerated
computing is accessible to every start-up, research institution, and enterprise.

Train, fine-tune, and serve AI at scale Python abstracted infrastructure

Developers and researchers can accelerate the Developers simply add a few lines of Python
entire AI development and deployment process code to access GPU and HPC infrastructure,
- from training, fine-tuning, benchmarking, data without the burden of learning Kubernetes,
synthesis, to inference - in a cost-effective way. Slurm, or other DevOps tools.

Simplify agentic workflows Run efficient AI & HPC infrastructure

Take AI agents to the next level by expressing Covalent automatically pools, assigns, and
complex interactions as a seamless Python scales resources to workloads dynamically, in
workflow and even deploying each agent in its order to meet changing demands for compute
own separate compute environment. and optimizing infrastructure efficiency.

Building an Environment Agnostic Workload Orchestration Platform with MongoDB

● Data Scalability: MongoDB Atlas provides ● Multi-Cloud Capability: MongoDB Atlas

an ideal foundation for modern AI and HPC allowed Covalent to reach multi-cloud
applications which require serverless compatibility faster than standard tooling
compute, autoscaling resources, and and give users the freedom to move data
distributed workloads between AWS, Google Cloud, and Microsoft
Azure.
● Ease of Use: MongoDB gives the small, agile
development team the freedom to build and ● Open Source Optionality: Covalent was
manage data workflows without the need for originally released as an open source project,
a specialist DBA and the MongoDB NoSQL database was the
perfect complement for quickly maturing
product roadmap.
Figure 54: Covalent by Agnostiq accelerates AI development and deployment with compute
orchestration, dynamic resource allocation, and auto-scaling infrastructure

Agnostiq chose MongoDB as their default MongoDB is used in the front-end, allowing a
NoSQL database for the free, open source high volume of metadata and other assets
version of Covalent. Without any DBAs as a to be published and cached in accordance
small agile team, MongoDB gave Agnostiq with an event-driven architecture. This near
the freedom to build and manage data real-time experience is key to a product
workflows without the need for a specialist. aimed at delivering a unified view over
distributed resources. MongoDB Atlas further
As their customer base grew along with the provides the autoscaling required to grow
demand for cloud computing access, with the user base and the number of
Agnostiq moved to MongoDB Atlas, gaining workloads while keeping costs in check.
the freedom to move data seamlessly
between AWS, Google Cloud, and Microsoft “MongoDB Atlas helps us provide an ideal
Azure. This gave Covalent the flexibility to foundation for modern HPC and AI
reach multi-cloud compatibility at a faster applications which require serverless
rate than with standard tooling. compute, autoscaling resources, distributed
workloads, and rapidly reconfigurable
Covalent provides a workflow management infrastructure.”
service by registering jobs, dispatching IDs, Santosh Kumar Radha, Head of Product at
and collecting other metadata that allows Agnostiq
fellow researchers and developers to
reproduce the original work.

© 2024 MongoDB, Inc. All rights reserved.

Credal.ai: Build secure AI
assistants for Enterprise
operations
Credal helps enterprises build secure AI applications: We provide a secure AI platform that
integrates seamlessly with internal data systems, ensuring privacy, compliance, and ease of use for
enterprises to build and manage AI applications.
Credal uses MongoDB Atlas for performant, high scale vector search for enabling AI powered
applications. Credal has multiple deployment options including Cloud, Managed Single Tenant and
On-prem with pluggable MongoDB Atlas instances managed either by Credal or the customer.

Give every employee the ability to create the perfect AI Assistant for their work.

Customer genie Product bot

Answer questions about any customer in a Slash time spent on internal knowledge sharing
shared Slack channel, from disparate data and back and forth, democratize product
sources. expertise.
Security questionnaires Sales coach
Speed up deals and customer time to value by Analyzes call transcripts & recommends on
offloading security approvals to AI. missed questions, discovery opportunities, and
objection handling.

Robust Security and Compliance for Your AI Initiatives

● Security & Compliance built in: Fully sync ● ROI from the SaaS tools your enterprise
permissions from source systems (eg Google relies on: Credal connects to the most
Drive, Atlassian products, and other SaaS important tools in your enterprise, such as
tools). Wide deployment options including Google Drive, Microsoft Sharepoint, Microsoft
Cloud, managed single tenant, cloud-prem, Onedrive, MongoDB, Box, Notion, Salesforce,
and on-prem to meet your security needs. Zendesk, Confluence, Jira and Slack.
Automatically redact sensitive data like Non-technical users can easily sync data
PII/PHI from AI to further improve from these tools for use in their assistants.
compliance and security posture, while Credal automatically honors the permissions
letting your users experiment, iterate and be on all your enterprise data - regardless of the
productive. Of course, Credal takes care of end user.
enterprise security fundamentals such as
● Trusted by Giants: Credal helps companies
SAML/SCIM, audit logging, and RBAC out of
scale AI and manage complex data, serving
the box.
clients like Wise, MongoDB, and unicorn
startups like Lattice.

MAAP Partner
© 2024 MongoDB, Inc. All rights reserved.
Figure 55: Bringing your data to LLMs and bring LLMs to your operations.

MongoDB integration Point and click AI Copilots

Credal seamlessly integrates your

Credal’s AI copilots empower users to set up
organization's MongoDB data with AI,
dedicated assistants for a wide range of use
offering direct developer access for flexible
cases, from customer support to contract
customization. It also supports dedicated
review.
collections for curated data with complex
retrieval strategies in MongoDB. Copilots can assist with any task that
Credal offers flexible deployment options: combines AI and data. Copilots are designed
cloud, managed single tenant, or to be experts on the data and context you
on-premises with pluggable MongoDB Atlas provide. They combine AI with your data to
instances. provide accurate, context-aware responses,
while citing their sources.
Leverage your existing investment in
MongoDB to integrate AI into your Copilots are useful to the full spectrum of
operations: Credal also lets developers and non-technical builders to highly skilled
data engineers use MongoDB as a data developers. Our API documentation is
source unto itself, to unlock new use cases available at docs.credal.ai.
that are only possible with your application
data.

© 2024 MongoDB, Inc. All rights reserved.

Anyscale enables MongoDB
users to deploy and seamlessly
scale their AI workloads
Anyscale, founded in 2019 by Professor Ion Stoica and the Berkeley RISELab team, developed Ray, a
leading AI Compute Engine for scaling AI workloads. The Anyscale Platform provides a
fully-managed version of Ray, optimized for performance, scalability, and developer collaboration.

Optimized Infrastructure & Scaling Leading Performance

Anyscale scales effortlessly across Anyscale’s Ray enhances performance,
heterogeneous clusters (GPUs/TPUs) on any scalability, cost-efficiency, and utilization
cloud, ensuring high reliability and optimized through optimizations ranging from hardware
performance for production workloads as data tweaks to intelligent instance management,
and user loads grow. making it the ideal platform for running Ray.
Any AI Workload, Any Framework Powerful Developer Tooling
Anyscale’s Ray Engine and high level libraries Anyscale Workspaces provide scalable
supports end-to-end AI workflows, from GenAI development environments, simplified
to deep learning, embeddings, and data dependency management, and powerful
processing. Use your data for any AI use case observability and debugging tools, streamlining
with any framework (TRT-LLM, vLLM, PyTorch, the dev-to-prod lifecycle so teams can build,
TensorFlow, etc.). test, and deploy faster.

Anyscale + MongoDB: A Powerful Pair for AI

● Seamless Integration:: MongoDB Atlas ● Gen AI and RAG Applications: With
integrates Anyscale’s AI Compute Platform Anyscale for fast and efficient LLM inference
effortlessly, enabling users to scale AI and MongoDB Atlas scalable vector indexing
workloads from Gen AI embeddings to model for contextual search, users can build and
training and inference. Both MongoDB Atlas deploy super scalable, AI-based
and Anyscale are available on AWS retrieval-augmented generation (RAG) flows
Marketplace for identical VPC deployments. and agentic systems..

● Scalable Efficient Multimodal Data ● Simplified Developer Experience:

Processing: Process and query diverse data MongoDB’s Developer Data Platform and
types, including images, text, and structured Anyscale’s distributed compute platform
data, in real-time, using MongoDB's flexible combine to give developers the power to
schema and Anyscale's scalable compute. deliver AI use cases across their organization
without managing infrastructure

MAAP Partner
© 2024 MongoDB, Inc. All rights reserved.
Figure 56: Multi-Modal Search Application - application is composed of multiple components -
including legacy search and the AI-enabled search - that work together to provide a hybrid search
experience

Multi-Modal Search is a cutting-edge use case that leverages Anyscale's distributed AI compute
engine and MongoDB Atlas's flexible NoSQL database to create a seamless search experience
across diverse data types, including text, images, and structured data. Anyscale provides the
infrastructure to easily scale and optimize compute resources, enabling rapid processing of
multimodal data, while MongoDB Atlas efficiently stores and indexes these datasets.

The combine solution makes it possible to perform fast and accurate similarity searches across
large volumes of complex, unstructured data. With this architecture, users can quickly retrieve
relevant content from a variety of formats, such as images, documents, and even video, without
being burdened by infrastructure management. This makes it ideal for building scalable, AI-driven
search applications across multiple domains while optimizing for resources utilization and
scalability.

● Enable vector search across multiple fields and dimensions via MongoDB Atlas Vector
Search

● Supporting the storage of multi-modal data like images, text, and structured data in
MongoDB Atlas

● Run performant LLM Batch Inference compute jobs with Anyscale’s Platform

● Enable highly available and scalable deployments with Anyscale Services

● Optimally scale and maximize utilization of costly compute resources with Anyscale’s
platform and intelligent infrastructure management

© 2024 MongoDB, Inc. All rights reserved.

Basikonʼs credit and leasing platform
that enables you to nurture your
customer, employee and partner
experiences
Basikon, a leader in financial technology, presents a powerful SaaS platform, transforming how
financial institutions oversee loans, leases, guarantees, and wholesale financing. Leveraging
cutting-edge technology and universal datacenters, Basikon processes millions of contracts daily,
providing innovative solutions for financial institutions to improve collaboration with partners and
customers.

Efficient Banking Modern Technology Utilization

The SaaS platform transforms financial The company uses modern tech to streamline
management, enabling a front-to-back loan processes, cut approval wait times, and prevent
and lease digital system in months, not years. errors, enhancing the customer experience with
automated end-to-end processes.

Orchestrating and Enabling Software Rapid Deployment

The platform orchestrates digital journeys The model enables rapid system deployment.
across front and back offices, managing Despite the complexity, over 15 systems have
financing product distribution directly or been deployed since 2019, setting it apart in
through partners, boosting agility and fintech.
productivity.

Building a Scalable and Agile Financial Services Platform

● Unmatched Scalability: Basikon leverages ● Seamless Cloud Integration: MongoDB's

MongoDB's horizontal scaling capabilities to compatibility with cloud platforms aligns
efficiently handle massive volumes of data, perfectly with Basikon's cloud-native
crucial for their lending and leasing solutions. approach, allowing for effortless integration
and operation.
● Streamlined Integrations: MongoDB's
robust API set simplifies integration with ● Adaptable Data Model: MongoDB's flexible
Basikon's pre-built integrations and external data model provides the agility to adapt to
systems, ensuring a smooth flow of data. ever-changing financial service requirements,
enabling Basikon to innovate and evolve
● High-Speed Operations: MongoDB's
rapidly.
exceptional performance on big data makes
it the ideal choice for Basikon, which
manages millions of contracts daily.
Figure 57: Basikon handling the complete customer life cycle

As the financing landscape evolves, there's a enables financial institutions to remain agile
growing need for configurable, cloud-based and responsive in today's competitive
software built on microservices. This type of landscape. This integrated approach not only
software should be able to manage the streamlines processes but also enhances
entire life cycle of any financing product. transparency and accountability throughout
Basikon, as illustrated in the Figure above, the financing journey, ultimately driving
fulfills this need by orchestrating the digital greater customer satisfaction and loyalty.
journey across all stages, from initial
customer interaction to loan approval and “MongoDB Atlas is very stable – in 4 years,
management. It also empowers financial we did not experience a single interruption
institutions to manage their distribution of service or find a single bug. Upgrades are
networks and partner relationships directly done in seconds with just the press of a
through the platform. button, increasing our agility 10x. At Basikon,
MongoDB has played a crucial role in our
Moreover, Basikon's cloud-based software success and we wouldn't be where we are
architecture built on microservices ensures today without it.”
adaptability to changing market dynamics Thomas Nokin, Founder and CEO at Basikon
and regulatory requirements. By offering a
comprehensive solution for managing the
entire financing product life cycle, Basikon

© 2024 MongoDB, Inc. All rights reserved.

Unlock Your Enterprise Content:
Encore's AI-Enabled Cloud
Solution Delivers Results
Remember all that messy, unavailable, unstructured data? Encore has turned that into a goldmine of
knowledge. With Encore’s AI-enabled platform, everything's ingested quickly, painlessly, and
organized to meet business needs. Encore is a SaaS, cloud-native, enterprise content management
platform which is a cost-effective solution for storing, retrieving & archiving business content.
Encore’s scalable repository leverages the power of MongoDB Atlas and the suite of AWS services
making finding content easy. Layering in AI-enabled services opens up the possibilities of
automation and efficiency to drive business growth.

AI-Enabled Services Service and Event-Driven Architecture

Organizations are sitting on a treasure trove of Transparency and ease of integration are
data in the form of documents, emails, images, central to the Encore design principles. Tracking
and other unstructured content. Encore enables all platform events published and making them
businesses to tap into this data using AI available through a suite of services provides
technology. insights that open up additional opportunities.

Enterprise Scale and Performance Secure, Complaint and Highly Available

Seamless horizontal scaling & redundancy Encore’s model puts security & compliance in
backed by the power and reliability of AWS & front of development & deployment. Their SOC2
MongoDB. The Encore platform scales to meet and StateRAMP compliance demonstrate their
whatever you need to optimally run your commitment to security excellence. DR/HA
business when you need it. removes any possibility of interruption.

Choice of MongoDB as Developer Data Platform

Flexible Schema: Encore’s is able to quickly Enterprise Scale: Demand is met with vertical
adapt to unique requirements. With MongoDB’s and horizontal scaling. The ease with which
flexible schema, they can add changes to the MongoDB provides scale-out to bring more
platform without creating complexity. nodes online or vertical scale to adjust
processing power is ideal for Encore customers.
Powerful Search: Businesses invest in content
solutions with an expectation that users can Ease of Use: MongoDB Atlas provides tools for
quickly locate documents and get accurate the product team to efficiently build new
answers to their open questions. MongoDB’s features and operationalize the Encore solution.
Vector capabilities provide the foundation for For example, aggregation pipelines for quick
all Semantic search and associated results to end users, or automatic failover in
embeddings. case of an outage.
Figure 58: The value of Encore

Optimizing your Business with AI-enabled Content

With the foundation of Encore built on prompts for the analysis. Encore’s crawlers
MongoDB Atlas, leveraging Semantic Search leverage AI and Vector Search to locate the
with Mongo’s Vector database has never documents that meet the criteria and create
been easier. Encore creates embeddings summaries to simplify the findings.
using the latest AI LLMs and stores them in
MongoDB Atlas, reimagining the search Vector Search: With Encore’s search,
capabilities on unstructured content. With powered by MongoDB's Vector Database
AI-enabled services, customers can gain and AWS Bedrock, finding content with
access to all the relevant data. simple user-provided prompts will retrieve
relevant content within seconds.
Document Summaries: Encore enables you
to simply press a single button to get a Workflow Automation: Leveraging the
summary of key points to help you quickly latest in AI technology with the Encore
gain insight into the content of interest. platform’s API architecture opens up the
opportunity for organizations to rethink
AI-Driven Analysis: Powering your expensive manual workflows. With Encore
organization with Encore translates to a you can eliminate manual steps that require
number of operational efficiency wins. a review of specific content and instead
Organizations are regularly tasked with automate searching for content, extract the
auditing archived content and expected to data from the content that is relevant to
respond quickly. With Encore AI-enabled your workflow criteria and expedite your
analysis, teams can use the collections customer requests.
feature to group content and provide

© 2024 MongoDB, Inc. All rights reserved.

How Cognigy Built a Leading
Conversational AI Solution With
MongoDB

Cognigy is a pioneering force in AI-driven customer service solutions on a global scale. They are at
the forefront of revolutionizing the customer service industry by providing the most cutting-edge AI
workforce on the market. Trusted by giants like Toyota, Bosch, and Lufthansa, their award-winning
solution empowers businesses to deliver exceptional customer service: instant, personalized, in any
language, and on any channel.

AI-Driven Customer Service Integration with Third-Party Platforms

Their main product, Cognigy.AI, allows Cognigy makes it simple to integrate with
companies to create AI Agents, improving third-party platforms like Facebook Messenger,
experiences through smart automation and Line, and WhatsApp. This broadens the reach of
natural language processing. This makes it easy customer service teams and helps businesses
for businesses to develop and deploy intelligent connect with their audience on various channels
voice and chatbots. they use.

Drag-and-Automate AI Enterprise-Level Security and Compliance

Cognigy's low-code platform lets business users Cognigy prioritizes security by offering features
build virtual agents with drag-and-drop tools that comply with industry standards like SOC 2,
like Flows, Playbooks, and Lexicons. GDPR, CCPA, and HIPAA.

Seamless Integration & Peak Performance

● MongoDB's JSON document storage aligns diverse data types, ensuring peak
perfectly with Cognigy's application performance under high loads.
language, facilitating seamless integration
● MongoDB empowered Cognigy.AI to handle
with Typescript and intuitive querying
expanding user interactions while
processes
maintaining peak performance, ensuring
● MongoDB's scalability via sharding aligns scalability and responsiveness in scaling
with Cognigy's growth vision, enabling conversational agents.
expansion across cloud providers and
● MongoDB's document model flexibility
on-premises setups.
enables easy data model modifications,
● MongoDB's developer data platform reducing concerns about data and schema
empowered Cognigy to efficiently manage migrations.
Figure 59: Cognigy’s replica-sets in production

Have you ever built a chatbot that struggled This tech allows Cognigy.AI to effortlessly
to keep up with user demands? Imagine a manage a growing number of user
platform that can handle hundreds of interactions, processing all sorts of data
queries per second, even during peak hours, easily.
all while storing massive amounts of data.
That's the power of MongoDB at work for Imagine your chatbot being able to learn
Cognigy.AI! and improve over time, This is what
MongoDB's flexible data model enables for
Cognigy constructed the platform by Cognigy.AI. As new data and user
employing a composable architecture model interactions flow in, Cognigy.AI can
with over 30 specialized microservices, which continuously update and refine its
they adeptly orchestrated through understanding of how to best serve your
Kubernetes. These microservices were customers. This collaboration is a prime
strategically fortified with MongoDB's replica example of how powerful technology can be
sets, spanning across three availability the driving force behind groundbreaking
zones, a move aimed at bolstering reliability products like Cognigy.AI. Imagine the
and fault tolerance. possibilities: chatbots that can provide
personalized recommendations, troubleshoot
As you can see in the Figure above, complex issues, and even have engaging
MongoDB's magic isn't just marketing hype. conversations.

© 2024 MongoDB, Inc. All rights reserved.

How Devnagri Brings the
Internet to 1.3 Billion People
with Machine Translations

Devnagri is India's leading AI-powered translation engine, enabling brands to localize content five
times faster and more accurately. As a SaaS platform, it focuses on translating Indian languages,
utilizing a hybrid approach of 80% machine and 20% human effort to achieve 99% accuracy in
translating millions of words daily.

Customizable AI models Tackling the digital divide in India

Devnagri trains machine translation models This AI platform offers machine translation for
with data stored in MongoDB Atlas, achieving non-English speakers, focusing on e-learning,
real-time translation. banking, e-commerce, and media.

Adapting to future advancements Human-in-the-loop approach

Devnagri considers using advanced models like Devnagri integrates human feedback to
OpenAI GPT-4 and Llama-2-7b, fine-tuned with enhance translation accuracy, emphasizing
their own translation data. their dedication to both automation and human
expertise for quality control.

Devnagri's Strategic Use of MongoDB for Machine Translation

● Flexible data model: MongoDB's document ● Access to expertise: Being part of
data model suits Devnagri's need to store MongoDB's AI Innovators Program grants
diverse structured and unstructured content Devnagri technical guidance and best
efficiently for training their machine practices, aiding their development process.
translation models.
● Scalability for performance: The
● Faster time to market: Efficiency aids distributed architecture of MongoDB allows
Devnagri in faster training and improved Devnagri to parallelize tasks across multiple
translation quality, accelerating product machines, improving training and translation
launches. speed.

● Supports real-time needs: Devnagri uses

data stored in MongoDB to train models for
real-time translation.
Figure 60: A visual of Devnagri’s real-time translation engine

Devnagri’s real-time translation engine helps over 100 Indian brands connect with their
customers over digital channels for the first time

The real-time translation engine has helped The platform's focus spans diverse industries
over 100 Indian brands connect with their such as e-learning, banking, e-commerce,
customers over digital channels for the first and media publishing, offering a tailored
time. This achievement signifies a solution beyond a general consumer tool.
breakthrough in overcoming the language Powered by custom transformer models and
barrier in India, where 90% of the population advancements like OpenAI GPT-4,
are not fluent in English, and more than 22 Devnagri's technology strives to democratize
Indian languages are in use. internet access for India's non-English
speakers.

© 2024 MongoDB, Inc. All rights reserved.

WINN.AI: The virtual assistant
tackling sales admin overhead

WINN.AI is more than just a tool; it’s a productivity powerhouse designed to transform the way sales
teams operate. By reducing administrative busywork, WINN.AI is helping organizations save time,
money, and resources, enabling sales teams to better invest their working hours in serving
customers.

AI-Powered Sales Assistant CRM Integration

An AI-powered real-time sales assistant joins After each meeting, WINN.AI extracts and
virtual meetings, understands conversation summarizes relevant information, updating the
context, and responds to customer queries, CRM system with follow-on actions, eliminating
enabling salespeople to focus on selling rather manual data entry, saving time, and reducing
than administrative tasks. errors.

Sales Playbook Prompts Contextual Understanding

WINN.AI can provide prompts from a sales The platform understands conversation context,
playbook, helping to guide the salesperson providing real-time relevant information to the
during customer interactions. It also ensures salesperson, including customer references and
meetings stay on track and on time. competitive data.

Building a Strong Foundation for AI

● Developer Familiarity: The developers at scaling, securing, and backing up their data,
WINN.AI are familiar with MongoDB, simplifying the tech stack and ensuring data
eliminating the need for database safety.
administrators or external experts and
● Cost Efficiency: By using MongoDB,
enabling the team to focus on building
WINN.AI can invest the savings from not
AI-powered products.
needing any DBA or external experts back
● Flexibility: MongoDB's flexibility allows into building great AI-powered products.
WINN.AI to handle data of any form,
● Stability: In the ever-changing AI tech
offering agility surpassing traditional
market, MongoDB serves as a stable anchor
relational databases.
for WINN.AI. This allows the developers to
● Managed Services: MongoDB Atlas provides freely create with AI while being able to
WINN.AI with managed services for running, maintain a reliable data infrastructure.
WINN.AI: The virtual assistant tackling sales admin overhead

Beyond simply attending meetings, WINN.AI fingertips during crucial customer

empowers salespeople by automating interactions.
tedious administrative tasks. After each
virtual encounter, WINN.AI intelligently Additionally, WINN.AI seamlessly integrates
summarizes key points and automatically with leading sales tools like Zoom, HubSpot,
updates the CRM system with follow-up and Salesforce, for a streamlined workflow.
actions. This eliminates the need for manual
data entry, saving salespeople valuable time
and minimizing errors.

Furthermore, WINN.AI boasts a powerful AI

architecture. Initially built on custom NLP
algorithms, the system now utilizes the
advanced capabilities of GPT 3.5 and 4 for
superior entity extraction and
summarization. This ensures salespeople
have the most relevant information at their

© 2024 MongoDB, Inc. All rights reserved.

Ada: Revolutionizing customer
service with AI-powered
automations built on MongoDB
Since 2016, Ada has become a dominant force in AI, reshaping customer service with its intelligent
automation engine. Their AI swiftly resolves complex inquiries across any channel, in any form.
Backed by nearly $200 million in funding and a team of 300 passionate innovators, Ada empowers
over 300 industry leaders, including tech titans like Meta, Verizon, and AT&T, to deliver exceptional
customer experiences.

AI-Powered Automations Rapid Product Development

Ada's advancements in transformer models, Ada prioritizes rapid product development,
LLMs, and RLHF have significantly enhanced measured by the speed of shipping products
their AI assistants, enabling advanced and features, as well as the pace of learning
reasoning to solve customer problems rather and iterating. They can deliver new products in
than just searching for information. just a few months.

Efficient Use of Unstructured Data Impressive Track Record

They can query unstructured data and use it to Since 2016, Ada has powered more than 4
train other models, enabling them to automate billion automated customer interactions for
queries and provide support that goes beyond brands like Wealthsimple, Verizon, AirAsia, Yeti,
just answering multi-step queries. and Square.

Unmatched Performance and Support: Keeping Ada Ahead

● Flexibility and Agility: Ada can easily scale ● Less Dependency on One Central Cloud
their database as their business grows and Vendor: By using a cloud-agnostic solution,
adapt to new channels and modalities Ada avoids being locked into a single cloud
without being restricted by their database provider. This gives them more freedom and
infrastructure. flexibility.

● Performance and Support: Ada has found ● Distributed Event Processing System: Ada
that the performance of MongoDB Atlas is using MongoDB Change Streams to build a
meets their needs, and they appreciate the distributed event processing system that
great support from the MongoDB team. powers bots and analytics.
Figure 61: Ada’s AI customer service

Beyond Automation, A Self-Learning AI for Superior Customer Service

Ada's focus on cutting-edge AI extends Furthermore, Ada prioritizes rapid
beyond just solving customer problems. They development, allowing them to deliver new
can automate tasks and provide advanced features and products in just a few months.
support by querying unstructured data, such This agility ensures they stay ahead of the
as customer conversations. This allows them curve in customer service innovation. In
to train additional models that go beyond essence, Ada is creating a self-learning AI
answering even complex, multi-step queries. loop that continuously improves customer
This translates to a superior customer service through automation and data-driven
experience as Ada can automate more insights.
interactions and provide more
comprehensive support.

© 2024 MongoDB, Inc. All rights reserved.

XOLTAR: Gen AI companion for
patient engagement and better
clinical outcomes

XOLTAR is a pioneering conversational AI platform designed to foster long-lasting patient

engagement. It provides an AI-powered accountability partner platform that mimics the one-on-one
interactions nurses conduct with patients. Through personalized encounters, these AI companions
guide patients toward adopting healthy habits necessary to manage their medical conditions.

AI Accountability Partner Platform Customizable AI Partners

Xoltar provides an AI accountability partner Each AI partner can be customized by gender,
platform that emulates the one-on-one race, and language, promote goals, monitor
interactions nurses conduct with patients. patients, collect and report video & audio RWE,
and offer an emotionally engaged experience.

Hyper-Personalized Encounters Sensor Fusion Technology

Through hyper-personalized encounters, the Xoltar’s sensor fusion technology interprets
accountability partners lead patients to human emotion from facial expressions, voice
embrace the healthy habits required to manage patterns, context, and other non-verbal cues.
their medical conditions.

Powering Patient Care with Real-Time Data and Machine Learning

● Long-term Memory and Model Training: ● Real-Time Interaction Management:

The data stored in MongoDB provides both XOLTAR can manage patient interactions in
long-term memory for each patient as well as real-time, thanks to the database. This is
input for ongoing model training and tuning. crucial for their omni-channel approach to
patient care.
● Event-Driven Data Pipelines: MongoDB
powers XOLTAR’s event-driven data pipelines. ● Real-Time Notifications: With Atlas
Follow-on actions generated from patient Triggers, MongoDB notifies downstream
interactions are persisted in MongoDB. consuming applications so they can react in
real-time to new treatment recommendations
● Support for Machine Learning Models:
and regimes.
MongoDB provides the necessary data for
training and fine-tuning XOLTAR's
sophisticated machine learning models.
Figure 62: How the XOLTAR platform works

At the Heart of XOLTAR lies a sophisticated data, such as video transcripts. This data
array of state-of-the-art machine learning provides both long-term memory for each
models working across multiple modalities — patient as well as input for ongoing model
voice and text, as well as vision for visual training and tuning.
perception of micro-expressions and
non-verbal communication. These custom MongoDB also powers XOLTAR’S
multilingual models are trained and event-driven data pipelines. Follow-on
deployed to create a truthful, grounded, and actions generated from patient interactions
aligned free-guided conversation, along with are persisted in MongoDB, with Atlas
various transformers for real-time automatic Triggers notifying downstream consuming
speech recognition. applications so they can react in real-time to
new treatment recommendations and
XOLTAR’s models personalize each patient’s regimes.
experience by retrieving data stored in
MongoDB Atlas. Taking advantage of the Through its participation in the MongoDB AI
flexible document model, XOLTAR developers Innovators program, XOLTAR’s development
store both structured data, such as patient team receives access to free Atlas credits
details and sensor measurements from and expert technical support, helping them
wearables, alongside unstructured de-risk new feature development.

© 2024 MongoDB, Inc. All rights reserved.

Conversation Intelligence with
Observe.AI

Observe.AI, a California-based company funded by over $200 million, is the leading provider of live
conversation intelligence for contact centers. Trusted by industry leaders like Accolade and Pearson,
Observe.AI empowers businesses to transform the way they interact with customers. The company is
focused on being the fastest way to boost contact center performance with live conversation
intelligence.

Advanced AI Techniques Efficient Operationalization

Observe.AI employs AI techniques, including Observe.AI optimizes MLOps with Docker and
transformers for NLP, for various tasks like text Kubernetes, enabling smooth model
classification, intent recognition, deployment, management, and scalability.
summarization, and question-answering.

Model Development and Training Speech Processing Expertise

Observe.AI uses TensorFlow and PyTorch to Observe.AI goes beyond NLP into speech
craft and fine-tune intricate natural language processing, using cutting-edge methods for
models, employing transfer learning and tasks like automatic speech recognition and
gradient-based optimization techniques. sentiment analysis to keep their language
capabilities leading-edge.

The role of MongoDB in Observe.AI technology stack

The MongoDB developer data platform gives because it enables us to quickly innovate,
the company’s developers and data scientists a scale to handle large and unpredictable
unified solution to build smarter AI workloads, and meet the security
applications.
requirements of our largest enterprise
customers.”
“OBSERVE.AI processes and runs models on Jithendra Vepa
millions of support touchpoints daily to Ph.D, Chief Scientist & India General Manager
generate insights for our customers. Most of at Observe.AI
this rich, unstructured data is stored in
MongoDB. We chose to build on MongoDB
Figure 63: Observe.AI’s conversation intelligence

Boost Sales and Support Teams: Data-Driven Insights from Observe.AI

The company has pioneered a 40 “Our products employ a versatile range of

billion-parameter contact center large AI and ML techniques, covering various
language model (LLM) and one of the domains. Within natural language
industry’s most accurate generative AI
processing (NLP), we rely on advanced
engines. Through these innovations,
algorithms and models such as
Observe.AI provides analysis and coaching
transformers, including the likes of
to maximize the performance of its
customers’ front-line support and sales
transformer-based in-house LLMs, for text
teams. classification, intent and entity recognition
tasks, summarization, question-answering,
Observe.AI's advanced AI tools analyze and more. We embrace supervised,
conversation data thoroughly, revealing key semi-supervised, and self-supervised
insights like emotions and sentiment. This
learning approaches to enhance our
helps businesses identify areas for
models' accuracy and adaptability."
improvement and provides targeted
coaching for exceptional customer service. Jithendra Vepa
Ph.D, Chief Scientist & India General
Manager at Observe.AI

© 2024 MongoDB, Inc. All rights reserved.

How Flagler Health's AI-Powered
Journey is Revolutionizing
Patient Care

Flagler Health is dedicated to supporting patients with chronic diseases by matching them with the
right physician for the right care. Typically, patients grappling with severe pain conditions face
limited options, often relying on prolonged opioid use or exploring costly and invasive surgical
interventions. Unfortunately, the latter approach is not only expensive but also has a long recovery
period. Flagler finds these patients and triages them to the appropriate specialist for an advanced
and comprehensive evaluation.

Flagler Health employs sophisticated AI techniques to rapidly process, synthesize, and analyze
patient health records to aid physicians in treating patients with advanced pain conditions. This
enables medical teams to make well-informed decisions, resulting in improved patient outcomes with
an accuracy rate exceeding 90% in identifying and diagnosing patients.

As the company built out its offerings, it identified the need to perform similarity searches across
patient records to match conditions. Flagler’s engineers identified the need for a vector database
but found standalone systems to be inefficient. They decided to use MongoDB Atlas Vector Search.

Creating an integrated platform to store all data in a single location with a unified
interface, facilitating quick access and efficient data querying.

● Flagler Health emphasized the importance of ● The managed services provided on

a flexible database that can evolve with MongoDB's developer data platform save
the company's growth. A relational model time and offer reliability at scale
was deemed too rigid, leading the company throughout the development cycle.
to choose MongoDB's document model.
● With Atlas Vector Search, developers can
● MongoDB’s flexibility allows for easy build AI-powered experiences while
customization of client configuration files, accessing all the data they need through a
streamlining data editing and evolution. unified and consistent developer
experience..
Figure 64: Current state without Flagler Health (left) and what Flagler Health can offer (right)

Flagler Health collaborates with many clinics, This comprehensive application architecture,
first processing millions of electronic health consolidated on MongoDB's developer data
record (EHR) files in Databricks and platform, simplifies Flagler Health's
transforming PDFs into raw text. Using the operations, enabling efficient development
MongoDB Spark Connector and Atlas Data and increased productivity. By preventing
Federation, the company seamlessly streams administrative loops, the platform ensures
data from AWS S3 to MongoDB. Combined timely access to potentially life-saving care
with the transformed data from Databricks, for patients.
Flagler’s real-time application data in
MongoDB is used to generate accurate and Looking ahead, Flagler Health aims to
personalized treatment plans for its users. enhance patient experiences by developing
MongoDB Atlas Search facilitates efficient new features, such as a digital portal
data search across Flagler Health's extensive offering virtual therapy and mental health
patient records. Beyond AI applications, services, treatment and recovery tracking,
MongoDB serves critical functions in Flagler and a repository of physical therapy videos.
Health's business, including its web Leveraging MongoDB’s AI Innovators
application and patient engagement suite, program for technical support and free Atlas
fostering seamless communication between credits, Flagler Health is rapidly integrating
patients and clinics. new AI-backed functionalities on the
MongoDB Atlas developer data platform to
further aid patients in need.

© 2024 MongoDB, Inc. All rights reserved.

Dataworkz: Generate Faster Data
Insights with Gen AI Apps &
Proprietary Data

The Dataworkz gen AI applications platform provides an all-in-one RAG as a Service to rapidly build,
deploy, operationalize and scale gen AI applications, and eliminates the complexity involved in
building reliable and scalable RAG applications. It includes advanced search and retrieval to provide
relevant context to LLMs, and monitoring with traceability to observe and optimize application
performance.

Visual RAG builder End-to-end traceability

No-code AI app development, with a An integrated, highly performant platform with
knowledge graph, and lexical and semantic comprehensive visibility into the underlying
search, plus frictionless data wrangling, to instrumentation and transactions.
create gen AI apps.

Composable AI stack Expand gen AI app adoption

Configure with your existing or new Implement additional use cases, connect new
technologies, with access to metrics, insights data sources and use RAG APIs to embed gen
and elastic deployment. AI in workflows easily, efficiently and securely.

MongoDB + Dataworkz | The Power of Combined Innovation

● For generative AI applications, Dataworkz ● Companies use the state-of-the-art

argues that a company's key differentiator, MongoDB Atlas technology to deliver their
or "superpower," lies in enhancing AI-enriched apps with the right security
underlying Large Language Models controls in place, and at the scale and
(LLMs) with its own well-managed data. performance users expect.

● To easily access diverse internal data in

MongoDB Atlas, Dataworkz is used—a
comprehensive RAG development platform.
Its Composable AI stack, hybrid search,
end-to-end traceability, and no-code data
transformation enhance gen AI applications.
Figure 65: A leading-edge platform to rapidly build, deploy, operationalize and scale gen AI
applications

Why Dataworkz: Unique capabilities for gen AI applications using RAG

Build Optimize
Visually Create Gen AI Applications: Comprehensive Visibility &
Develop gen AI applications using a visual Customization: Achieve complete AI stack
RAG builder. This eliminates the need to transparency and easily customize data
worry about the complexity of the underlying processing steps in a user-friendly, no-code
infrastructure. interface.

Smart Routing with Knowledge Graph: Data-Driven Optimization: Conduct A/B

Set up smart routing using a knowledge testing on RAG pipelines using built-in
graph for lexical and semantic search. evaluation metrics to determine the most
effective configurations.

Observe Scale
End-to-End Traceability: Get full visibility Rapid Application Development: Build
of your gen AI apps with end-to-end diverse gen AI applications efficiently by
traceability for better performance utilizing pre-defined templates for various
optimization. use cases.

Centralized Monitoring: Track all system Embeddable RAG: Integrate gen AI apps
activity, like LLM calls, SLM calls, indexing, with Slack, Azure Studio, and HTML widgets
and retrieval, with one unified tool. via RAG APIs for enhanced accessibility in
workflows.

© 2024 MongoDB, Inc. All rights reserved.

VISO TRUST: Transforming
cyber risk intelligence

VISO TRUST is an AI-powered platform that helps companies quickly assess the cybersecurity risk of
their vendors. It provides actionable security information in minutes, allowing businesses to make
informed decisions with ease. VISO TRUST boasts a 90% reduction in workload and an 80% faster
risk assessment process, with near-universal vendor adoption by their clients.

Automated Risk Management Artifact Intelligence

VISO TRUST uses AI to streamline third-party Curated AI extracts insights from source
risk assessments, enabling instant evaluation artifacts, automatically determining vendor
without extra analysts. It eliminates lengthy security posture. This frictionless due diligence
questionnaires and manual document analysis process simplifies assessing any number of
for a more efficient approach. third parties.

Risk Insights Compliance Excellence

On the platform, users can gain a Continuously exceeding ISO, NIST, AICPA, and
comprehensive overview of their organization’s other standards without impeding business
cyber risk posture, enabling them to make operations is made possible by VISO TRUST. It
data-driven decisions to reduce risk across all empowers organizations to take control of their
third-party relationships. third-party security posture.

Empowering Customers with Faster Insights

● VISO TRUST deploys discriminator models ● The outcomes of RAG serve as the
that produce high-confidence predictions foundation for seeding LLM prompts and
about features of the artifact. linking their outputs in a chain, resulting in
the generation of highly precise factual
● The artifacts undergo a process where their
details regarding the artifact in the pipeline.
text content is extracted and integrated into
This data facilitates the swift delivery of
MongoDB Atlas, thus becoming integrated
intelligence to customers, a task that
into the dense retrieval system. This system
previously required weeks to accomplish.
executes Retrieval-Augmented Generation
(RAG) by leveraging MongoDB functionalities
such as Atlas Vector Search. Its aim is to
furnish ranked context to prompts for large
language models (LLMs).
Figure 66: Insights dashboard

Streamlining Third-Party Cyber Risk Management

VISO TRUST is the only SaaS third-party VISO TRUST uses state-of-the-art models
cyber risk management platform that from OpenAI, Hugging Face, Anthropic,
delivers the rapid security intelligence Google, and AWS, augmented by vector
needed for modern companies to make search and retrieval from MongoDB Atlas.
critical risk decisions early in the Read our interview blog post with VISO
procurement process. TRUST to learn more.

© 2024 MongoDB, Inc. All rights reserved.

How DevRev is Redefining CRM
for Product-Led Growth

OneCRM from DevRev is purpose-built for Software-as-a-Service (SaaS) companies. It brings

together previously separate customer relationship management (CRM) suites for product
management, support, and software development. Built on a foundation of customizable large
language models (LLMs), data engineering, analytics, and MongoDB Atlas, it connects end users,
sellers, support, product owners, and developers. OneCRM converges multiple discrete business apps
and teams onto a common platform.

The multi-cloud architecture of Atlas provides flexibility and choice that proprietary offerings from
the hyperscalers can’t match. While DevRev today runs on AWS, in the early days of the company,
they evaluated multiple cloud vendors. Knowing that MongoDB Atlas could run anywhere gave them
the confidence to make a choice on the platform, knowing they would not be locked into that choice
in the future.

DevRev manages critical customer data, and so relies on MongoDB Atlas’ native encryption and
backup for data protection and regulatory compliance. The ability to provide multi-region databases
in Atlas means global customers get further control over data residency, latency, and high
availability requirements.

CRM + AI: Digging into the stack

DevRev’s Support and Product CRM serve over This data is also encoded by open-source
4,500 customers: embedding models where it is used alongside
OpenAI models for customer support chatbots
● Support CRM brings support staff, product
and question-answering tasks orchestrated by
managers, and developers onto an AI-native
autonomous agents. MongoDB partner
platform to automate Level 1 (L1), assist L2,
LangChain is used to call the models, while also
and elevate L3 to become true collaborators.
providing a layer of abstraction that frees
● Product CRM brings product planning, DevRev engineers to effortlessly switch
software work management, and product between different generative AI models as
360 together so product teams can needed.
assimilate the voice of the customer in
Data flows across DevRev’s distributed
real-time.
microservices estate and into its AI models are
AI is central to both the Support and Product powered by MongoDB change streams.
CRMs. The company’s engineers build and run Downstream services are notified in real-time
their own neural networks, fine-tuned with of any data changes using a fully reactive,
application data managed by MongoDB Atlas. event-driven architecture.
Figure 67: Event-driven microservices architecture for DevRev’s AI-powered CRM platform

MongoDB Atlas: AI-powered CRM on an agile and trusted data platform

MongoDB is the primary database backing and release faster. Developers can
OneCRM, managing users, customer and experiment locally, then move to integration
product data, tickets, and more. DevRev testing, and then production — all running in
selected MongoDB Atlas from the very different environments — without changing
outset of the company. The flexibility of its a single line of code. This is core to DevRev’s
data model, freedom to run anywhere,
velocity in handling over 4,000 pull requests
reliability and compliance, and operational
per month:
efficiency of the Atlas managed service all
impact how quickly DevRev can build and
● Developers can experiment and test with
ship high-quality features to its customers.
MongoDB on local instances — for
The flexibility of the document data model example adding indexes or evaluating
enables DevRev’s engineers to handle the new query operators, enabling them to
massive variety of data structures their catch issues earlier in the development
microservices need to work with. Documents cycle.
are large, and each can have many custom ● Once unit tests are complete, developers
fields. To efficiently store, index, and query can move to temporary instances in
this data, developers use MongoDB’s Docker containers for end-to-end
Attribute pattern and have the flexibility to
integration testing.
add, modify, and remove fields at any time.
● When ready, teams can deploy to
The freedom to run MongoDB anywhere production in MongoDB Atlas.
helps the engineering team develop, test,

© 2024 MongoDB, Inc. All rights reserved.

 Elevating the edge experience:
Deploy AI anywhere with
Cloneable and MongoDB
Cloneable provides the application layer that brings AI to any device at the edge of the network. The
Cloneable platform empowers developers to craft dynamic applications using intuitive low/no-code
tools, instantly deployable to a spectrum of devices - mobiles, IoT devices, robots, and beyond.

Component-Based Development AI Object Detection

Cloneable apps are built using components, Cloneable provides an AI model for object
ranging from simple logic to complex data detection. You can process input images from
processing. These components allow you to video previews or captured photos, and the
construct applications that solve real-world model detects objects, outputting bounding
problems by layering them together in the app boxes and relevant statistics based on business
builder. rules.

Augmented Reality (AR) GIS Mapping

Cloneable's AR component, empowers users to Cloneable leverages ESRI technology to enable
interact with field assets in real time. Whether smart, data-driven mapping styles. With
navigating to a specific location or identifying intuitive analysis tools, you can gain location
an asset for inspection, AR enhances the user intelligence across field assets.
experience.

Real-Time Operational Tracking and Analysis

● Cloneable integrates seamlessly with ● Utilizing Cloneable and Atlas Vector Search
MongoDB, enabling the persistence of data to generate vector embeddings from images
locally on devices and its synchronized and device data enables users to efficiently
transfer to the cloud-based Atlas database. search and analyze field-collected events,
This ensures that enterprises can track, thereby enhancing decision-making and
measure, and respond to events across insights.
their operations in real-time.
Figure 68: Cloneable components

Empowering Businesses with Efficiency, Personalization, and Growth

By harnessing machine learning models, a personalization and customer engagement.

business can seamlessly leverage complex These models enable companies to analyze
technologies across its operations. Models vast amounts of data to understand
are pushed down to the device where they customer behavior, preferences, and trends,
are converted to a native embedded format allowing for tailored recommendations,
such as CoreML. From here, they are targeted marketing campaigns, and
executed by the device’s neural engine to interactive experiences. By leveraging
provide low latency inference, computer machine learning in this way, businesses can
vision, and augmented reality. forge deeper connections with their
customers, leading to increased satisfaction,
In addition to the operational efficiency loyalty, and ultimately, improved business
gained through machine learning models, outcomes.
businesses also benefit from enhanced

© 2024 MongoDB, Inc. All rights reserved.

 How Patronus Automates LLM
Evaluation to Boost Confidence
in Gen AI

Patronus AI is a company that develops tools to help businesses safely use large language models
(LLMs). Their main product is an automated evaluation platform that can identify errors and
unreliable outputs from LLMs. This is especially important for regulated industries where mistakes
can have serious consequences.

Founded by machine learning experts from Meta AI and Meta Reality Labs, Patronus AI is on a
mission to boost enterprise confidence in gen AI-powered apps, leading the way in shaping a
trustworthy AI landscape.

“Our platform enables engineers to score and benchmark LLM performance on

real-world scenarios, generate adversarial test cases, monitor hallucinations, and detect
PII and other unexpected and unsafe behavior. Customers use Patronus AI to detect LLM
mistakes at scale and deploy AI products safely and confidently.”
Rebecca Qian
Co-founder and CTO at Patronus

Overcoming LLM hallucination

In recently published and widely cited research While retrieval augmented generation (RAG) is
based on the FinanceBench question answering a common way of feeding models with
(QA) evaluation suite, Patronus made a up-to-date, domain-specific context, a key
startling discovery. Researchers found that a question faced by app owners is how to test the
range of widely used state-of-the-art LLMs reliability of model outputs in a scalable way.
frequently hallucinated, incorrectly answering or This is where Patronus comes in. The company
refusing to answer up to 81% of financial has partnered with the leading technologies in
analysts’ questions! This error rate occurred the gen AI ecosystem — from model providers
despite the models’ context windows being and frameworks to vector store and RAG
augmented with context retrieved from an solutions — to provide managed evaluation
external vector store. services, test suites, and adversarial data sets.
Figure 69: Reference architecture and workflow

Boosting confidence in LLMs with MongoDB

As Patronus assessed the landscape to ● The LlamaIndex data framework to
prioritize which partners to work with, they ingest and chunk the source pdf
saw massive demand from customers for document
MongoDB Atlas. Through the Patronus RAG ● Atlas Vector Search to store, index,
evaluation API, they help customers verify and query the chunk’s metadata and
that their RAG systems built on top of embeddings
MongoDB Atlas consistently deliver top-tier, ● Patronus to score the model
dependable information. responses

In its new 10-minute guide, Patronus takes Equipped with the results of an analysis,
developers through a workflow showcasing there are a number of steps developers can
how to evaluate a MongoDB Atlas-based take to improve the performance of a RAG
retrieval system. The guide focuses on system. These include exploring different
evaluating hallucination and answers indexes, modifying document chunking sizes,
relevance against an SEC 10-K filing, re-engineering prompts, and for the most
simulating a financial analyst querying the domain-specific apps, fine-tuning the
document for analysis and insights. The embedding model itself. Review the
workflow is built using: 10-minute guide for a more detailed
explanation of each of these steps.

© 2024 MongoDB, Inc. All rights reserved.

How Gradient Accelerator Blocks
Take You From Zero To AI in
Seconds
Gradient, founded by AI experts from Google, Netflix, and Splunk, helps businesses build
high-performing, cost-effective custom AI applications. It provides a platform for businesses to
build, customize, and deploy bespoke AI solutions — starting with the fastest way to develop AI
through the use of its Accelerator Blocks.

Fast Development with Pre-built Blocks Benefits for Regulated Industries

Gradient offers Accelerator Blocks - pre-built The platform empowers regulated industries
solutions for common AI tasks like entity such as finance and healthcare businesses with
extraction or document summarization. These data and AI control for regulatory compliance,
blocks can be used directly or combined for offering industry-specific models and
more complex needs, reducing development performance/cost benefits.
time and effort.

“With MongoDB, developers can store data of any structure and then expose that data to OLTP, text search,
and vector search processing using a single query API and driver. With this unification, developers have all
of the core data services they need to build AI-powered apps that rely on working with live, operational
data.”
Tiffany Peng, VP of Engineering at Gradient

Simplified RAG with Powerful Tech

● Simplified Infrastructure: Gradient’s ● operational databases and vector search

Accelerator Block for retrieval augmented capabilities in a unified, fully managed
generation (RAG) leverages MongoDB Atlas solution.
Vector Search and LlamaIndex. By using these
● Seamless Data Handling: With MongoDB,
technologies, Gradient eliminates the need for
developers can store data of any structure
complex infrastructure setup or deep
and expose it to OLTP, text search, and
knowledge of retrieval architectures.
vector search processing using a single query
● Best-of-Breed Technologies: Gradient API and driver. This unification provides all
partners with key vendors and communities in the core data services needed to build
the AI ecosystem. MongoDB Atlas, included as AI-powered apps that work with live,
a core part of the Gradient platform, provides operational data.
Figure 70: Managed RAG service

Gradient's Accelerator Block Boosts RAG Model Performance and Accuracy with
Pre-Built Infrastructure
Gradient’s newest Accelerator Block focuses Together, Atlas Vector Search and
on enhancing the performance and accuracy LlamaIndex feed foundation models with
of a model through retrieval augmented up-to-date, proprietary enterprise data in
generation (RAG). The Accelerator Block uses real-time. Gradient designed the Accelerator
Gradient’s state-of-the-art LLMs and Block for RAG to improve development
embeddings, MongoDB Atlas Vector Search velocity up to 10x by removing the need for
for storing, indexing, and retrieving infrastructure, setup, or in-depth knowledge
high-dimensional vector data, and around retrieval architectures. It also
LlamaIndex for data integration. incorporates best practices in document
chunking, re-rankers, and advanced retrieval
strategies.

© 2024 MongoDB, Inc. All rights reserved.

One AI: Providing AI-as-a-Service
to deliver solutions in days rather
than months

One AI is a company that aims to democratize and deliver AI as a service for businesses. Their
mission is to integrate AI into everyday life by transforming natural language into structured,
actionable data. This is achieved through their easy-to-use APIs, which package leading AI
capabilities from across the ecosystem.

AI-as-a-Service API’s for Developers

One AI provides AI-as-a-Service, delivering The One AI APIs allow developers to analyze,
solutions in days rather than months. This process, and transform language input in their
allows businesses to deploy tailored AI code, without requiring any training data or
solutions quickly and efficiently. NLP/ML knowledge.

Diverse Use Cases Flexible Data Infrastructure

One AI’s customers span multiple domains, One AI works with over 20 different AI/ML
utilizing their service for a variety of use cases, models and leverages a flexible data
from analyzing financial documents to infrastructure, specifically the MongoDB
AI-automated video editing. document model, to continuously explore and
add new capabilities for the AI.

Choice of MongoDB as Developer Data Platform

● Focus on Core Mission: MongoDB allows ● Flexible Data Infrastructure: With

One AI to focus on their core mission of using MongoDB, One AI can add, expand, and
AI to derive meaning from large volumes of explore new capabilities on a continuous
unstructured text1. Dealing with database basis.
requirements and services, such as
● Regular New Releases: One AI benefits
managing the pipeline, storage, and
from regular new releases from MongoDB,
backups, involves a lot of time, effort, and
such as Atlas Vector Search. This feature
hassle. MongoDB handles these tasks,
allows One AI to have vectorized language
allowing One AI to concentrate on their main
representation in the same database as
objective.
other representations, which can be accessed
via a single query interface. This solves a core
problem for One AI as an API company.
Figure 71: The One AI Language Studio

The One AI APIs let developers analyze, process, and transform language input in their
code. No training data or NLP/ML knowledge are required.

“The MongoDB document model really allows The company also benefits from the regular
us to spread our wings and freely explore new new releases from MongoDB, such as Atlas
capabilities for the AI, such as new predictions, Vector Search, which Ben sees as a highly
new insights, and new output data points.” Ben valuable addition to the platform’s toolkit.
adds, “With any other platform, we would have Ben explains: “The ability to have that
to constantly go back to the underlying vectorized language representation in the
infrastructure and maintain it. Now, we can same database as other representations,
add, expand, and explore new capabilities on a which you can then access via a single query
continuous basis.” interface, solves a core problem for us as an
API company."
Amit Ben, CEO at One AI

© 2024 MongoDB, Inc. All rights reserved.

Kovai: Bringing the power of
Vector Search to enterprise
knowledge bases

Founded in 2011, Kovai is an enterprise software company that offers multiple products in both the
enterprise and B2B SaaS arena. Since its founding, the company has grown to nearly 300 employees
serving over 2,500 customers.

Document 360 AI Assistant “Eddy”

Kovai’s key product, Document360, is a Kovai recognized the growing importance of AI
knowledge base platform designed for SaaS and developed an AI assistant named “Eddy”.
companies seeking a self-service software Eddy leverages LLMs (Language Models) and
documentation solution. It enables efficient retrieves information from the Document360
management and sharing of critical knowledge base to provide accurate answers to
information. customer queries.

“Atlas Vector Search is robust, cost-effective, and blazingly fast!”

Said Saravana Kumar, CEO, Kovai, when speaking about his team's experience

Choice of MongoDB as Developer Data Platform

● MongoDB Vector Search offers architectural ● Atlas Vector Search enables Kovai to store
simplicity, making it easier for Kovai to both knowledge base articles and their
optimize the technical architecture needed to embeddings together in MongoDB
implement their AI assistant, "Eddy." This collections. This eliminates the need for data
simplicity likely streamlines development syncing between multiple databases, which
efforts and reduces complexity in not only simplifies operations but also
integrating the search functionality into their reduces potential inaccuracies in answers
system. provided by the assistant. Operational
efficiency is crucial for a seamless user
● MongoDB Vector Search delivers faster
experience.
query response times at scale, ensuring a
positive user experience for Kovai's
customers interacting with the AI assistant.
Figure 72: Reference architecture

Faster, Simpler, More Efficient: How Kovai Leverages MongoDB Atlas Vector Search

The release of MongoDB Atlas Vector Search Specifically, the team has seen the average
provided a solution with three key time taken to return three, five, and 10
advantages for the engineers: chunks between two and four milliseconds,

● Architectural simplicity: MongoDB and if the question is a closed loop, the

Vector Search's architectural average time reduces to less than two
simplicity helps Kovai optimize the milliseconds.
technical architecture needed to
implement Eddy.
● Operational efficiency: Atlas Vector
Search allows Kovai to store both
knowledge base articles and their
embeddings together in MongoDB
collections, eliminating “data
syncing” issues that come with other
vendors.
● Performance: Kovai gets faster
query response from MongoDB
Vector Search at scale to ensure a
positive user experience.

© 2024 MongoDB, Inc. All rights reserved.

Robust Intelligence: Securing
generative AI, supercharged by
your data

Robust Intelligence safeguards organizations from AI's risks. Their end-to-end platform continuously
validates models, protecting them with an AI Firewall. This empowers confident AI adoption for any
model type, from basic to generative. Trusted by leaders like JPMorgan Chase, Robust Intelligence is
your key to unlocking AI's potential.

Recent advancements in generative AI have motivated companies to experiment with potential

applications, but a lack of security controls has exposed companies to unmanaged risks. This
challenge is exacerbated when sensitive company information is used to enrich pre-trained models,
such as connecting vector databases, in order to increase the relevance to the end user.

Robust Intelligence's AI Firewall safeguards large language models (LLMs) in production by

validating inputs and outputs in real-time. It addresses operational risks like hallucinations, ethical
risks such as model bias and toxic outputs, and security risks like prompt injections and PII
extraction. By intercepting harmful inputs and filtering out undesirable AI-generated outcomes, the
AI Firewall ensures model integrity and application safety.

“By incorporating MongoDBʼs Atlas Vector Search into the AI validation process,
customers can confidently use their databases to enhance LLM responses
knowing that sensitive information will remain secure. The integration provides
seamless protection against a comprehensive set of security, ethical, and
operational risks.”
Yaron Singer
CEO and co-founder at Robus Intelligence
Figure 73: High level architecture

Unlocking Personalized Customer Experiences with Algomo's Conversational AI

Customers can confidently connect Additionally, it serves as the memory and

MongoDB Atlas Vector Search to any database to store historical data points. This
commercial or open-source LLM for secure is important in the context of identifying
retrieval-augmented generation with the AI more advanced security attacks, such as
Firewall integration. Atlas Vector Search data poisoning and model extraction, which
serves as the memory and fact database for often manifest across a cluster of data
AI Firewall, ensuring the AI model provides points as opposed to a single data point.
enriched responses without hallucinating.

© 2024 MongoDB, Inc. All rights reserved.

Component-Based AI
for Development
Teams

Solutions built out of building

blocks can be seamlessly
integrated into existing systems
without disrupting other
functions or data
Fireworks AI and MongoDB: The
Fastest AI Apps with the Best
Models, Powered By Your Data

Fireworks AI and MongoDB are now partnering to make innovating with generative AI faster, more
efficient, and more secure. Fireworks AI was founded in late 2022 by industry veterans from Meta’s
PyTorch team, where they focused on performance optimization, improving the developer
experience, and running AI apps at scale. It’s this expertise that Fireworks AI brings to its production
AI platform, curating and optimizing the industry's leading open models. Benchmarking by the
company shows gen AI models running on Fireworks AI deliver up to 4x faster inference speeds than
alternative platforms, with up to 8x higher throughput and scale.

Models are one part of the application stack. But for developers to unlock the power of gen AI, they
also need to bring enterprise data to those models. That’s why Fireworks AI has partnered with
MongoDB, addressing one of the toughest challenges to adopting AI. With MongoDB Atlas,
developers can securely unify operational data, unstructured data, and vector embeddings to safely
build consistent, correct, and differentiated AI applications and experiences. Fireworks AI and
MongoDB provide a solution for developers who want to leverage highly curated and optimized
open-source models, and combine these with their organization’s own proprietary data — and to do
it all with unparalleled speed and security.

Lightning-fast models from Fireworks AI: Enabling speed, efficiency, and value

With its lightning-fast inference platform, ingest JSON-formatted objects from

Fireworks AI curates, optimizes, and deploys databases such as MongoDB Atlas.
40+ different AI models, resulting in significant
● Simple interfaces and APIs for
cost savings, reduced latency, and improved
development and production: The
throughput. Their platform delivers this via:
Fireworks AI playground allows developers to
● Off-the-shelf models, optimized models, interact with models right in a browser. It can
and add-ons: Fireworks AI provides a also be accessed programmatically via a
collection of top-quality text, embedding, convenient REST API. This is OpenAI
and image foundation models. Developers API-compatible and thus interoperates with
can leverage these models or fine-tune and the broader LLM ecosystem.
deploy their own, pairing them with their own
● Cookbook: A simple and easy-to-use
proprietary data using MongoDB Atlas.
cookbook provides a comprehensive set of
● Fine-tuning capabilities: To further improve ready-to-use recipes that can be adapted for
model accuracy and speed, Fireworks AI also various use cases, including fine-tuning,
offers a fine-tuning service using its CLI to generation, and evaluation.

MAAP Partner
Figure 74: Bringing your data to LLMs

Getting started: The Fireworks tutorial showcases how to bring your own data to LLMs
with retrieval-augmented generation (RAG) and MongoDB Atlas

With Fireworks AI and MongoDB Atlas, apps provide better intelligence and strategies.
run in isolated environments ensuring uptime Or, organize and classify a product
and privacy, protected by sophisticated catalog using product images and text.
security controls that meet the toughest
● Images to structured data extraction:
regulatory standards:
Extract meaning from images to produce
● As one of the top open-source model API structured data that can be processed
providers, Fireworks AI serves 140 billion and searched in a range of vision apps —
tokens per day (and growing). from stock photos, to fashion, to object
detection, to medical diagnostics.
● With Atlas, you run your apps on a proven
platform that serves tens of thousands of ● Alert intelligence: Process large amounts
customers, from high-growth startups to of data in real-time to automatically
the largest enterprises and governments. detect and alert on instances of fraud,
cybersecurity threats, and more.
Together, the Fireworks AI and MongoDB
joint solution enables: Getting started with Fireworks AI and
● Retrieval-augmented generation (RAG) MongoDB Atlas: review the Optimizing RAG
or Q&A from a vast pool of documents: with MongoDB Atlas and Fireworks AI
Ingest a large number of documents to tutorial, which shows you how to build a
produce summaries and structured data
that can then power conversational AI. movie recommendation app.

● Classification through
semantic/similarity search: Classify and
analyze concepts and emotions from sales
calls, video conferences, and more to

© 2024 MongoDB, Inc. All rights reserved.

LangChain: build, test, and
monitor enterprise-ready LLM
applications and agents

LangChain and LangGraph are open-source frameworks for building context-aware reasoning
applications and reliable agents. LangSmith is a unified developer platform for building, testing, and
monitoring LLM applications, whether you use LangChain or not.

Join 1M+ builders who standardize their development using LangChain’s frameworks and platform
together or separately to accelerate their AI application development.

Flexibility Develop, Debug, and Monitor

Build context-aware, reasoning applications With LangSmith, see what your agent or chain is
with LangChain’s flexible framework that doing and what context is being provided to the
leverages your company’s data and APIs prompts. See what’s happening in production.

Reliability Evaluate and Test

Use LangGraph as your controllable agent Layer in human feedback on LangSmith runs or
orchestration framework for handling complex use automatic evaluation. Stress-test quality
tasks reliably. over large datasets.

AI Efficiency, Simplified: LangChain and MongoDB

● Boosting Retrieval Performance: With ● Enhanced Search: The LangChain

MongoDB's efficient data storage and integration makes it easy to vectorize your
LangChain's flexible frameworks, you can operational data from and to MongoDB
increase operational productivity, improve Atlas, allowing for more efficient and
discovery and personalization, and enhance accurate retrieval of documents, even when
scalability and performance of your LLM the queries are complex or ambiguous, which
applications. can significantly improve the performance
and accuracy of search applications.
● Creating Tailored Solutions: MongoDB's
vector database and LangChain's application
framework make it possible to build and
deploy solutions that leverage your
proprietary data to address many use cases.

MAAP Partner
© 2024 MongoDB, Inc. All rights reserved.
Augment the power of LLMs with your data

The most powerful LLM applications and Systems composing LangChain, MongoDB,
agents integrate with public data (via LLMs and an LLM together enable you to extract
and search engines), external systems (via valuable insights from complex and recent
tools), and most importantly, your data, highlighting the potential of these
organization’s private documents and data. technologies in developing intelligent
applications. For more information, please
MongoDB’s deep integration with
refer to this detailed blog.
LangChain’s Python and TypeScript
frameworks provides the powerful semantic
search capabilities of MongoDB Atlas’s
vector search engine over your private data.

Start with a retrieval-augmented generation

workflow just over your private data, expand
your functionality with real-time search
engine powered features, and finally give
LangGraph agents access to your APIs to
interface with other systems.

© 2024 MongoDB, Inc. All rights reserved.

Build generative AI applications
over your enterprise data with
LlamaIndex and MongoDB
LlamaIndex is a developer platform that accelerates building agentic applications over your
enterprise data. It has two main products: a development framework in Python/Typescript that
is one of the most popular orchestration tools for building agentic applications. Another is
LlamaCloud, which provides an end-to-end RAG platform freeing up dev teams to focus on
business facing logic. LlamaCloud includes LlamaParse, an industry-leading document parsing
solution that handles complex documents, formatting them to be easily understood by LLMs.

Accelerated time to production

LlamaIndex exables 5-10x faster development, from start to production deployment, reducing
development costs (e.g. 2 engineers delivering production apps in a few weeks), and enabling rapid
iteration on multiple use cases.

Scalable and Enterprise ready Higher performance and accuracy

LlamaIndex can scale to handle large volumes Customers see a significant boost in retrieval
of enterprise data. It provides enterprise grade performance, as much as 2x over naive retrieval
security, including in VPC deployments over complex documents (e.g. Docs with Tables,
Charts, Figures)

LlamaIndex and MongoDB Atlas - better together

● Seamless Integration: MongoDB Atlas as a ● Scalable Applications: Customers can build

Data Sink (VectorDB) is supported in AI-driven applications, such as agents or
LlamaCloud and LlamaIndex framework. This assistants, with scalability and flexibility,
works for SaaS as well as in VPC benefiting from MongoDB's robust data
deployments and takes just a few clicks. management and LlamaIndex’s enterprise
ready offerings
● High performance results: Atlas Vector
Search Index is automatically created and
vector search queries can be run via the
advance retrieval algorithms provided by
LlamaIndex!

MAAP Partner
© 2024 MongoDB, Inc. All rights reserved.
Figure 75: LlamaCloud indexes data from various sources and enables advanced retrieval
algorithms. This can be connected to the open-source LlamaIndex for agentic workflows, Q&A,
chatbot etc

LlamaIndex provides the fastest way to build production-ready AI applications over

your company's data.

LlamaCloud LlamaIndex (open source framework)

● Enterprise RAG platform: Connect
with your data sources and provide ● Comprehensive AI Development
advanced multimodal RAG Framework to build Advanced AI
capabilities Applications
● Rapid Deployment and Performance:
Set up data pipelines for LLMs in a ● Seamless Integrations with data
few clicks, while minimizing sources, LLMs, and vectorDBs along
hallucination and ensuring high with customizability
retrieval performance
● Security and Scalability: Enterprise ● Strong community and ecosystem
grade security and scalable
infrastructure for all deployment
sizes

© 2024 MongoDB, Inc. All rights reserved.

Nomic AI: Cost-effective, Open
Source Embeddings at Scale

Nomic Embed v1.5 is a truly open source text embedder for the big data era. Out of the box, this
model supports a 8192 token context length, resizable embedding dimensions, and binary
quantization, all while outperforming similar models such as OpenAI’s Ada-002 and
text-embedding-3-small on both short and long context tasks.
Truly Open Source Long Context
Nomic Embed provides open-source model Nomic Embed supports a 8192 token context
weights and training code under the Apache-2 length making it well-suited for real-world
license, with curated training data available on applications with large PDFs and text
the Nomic website. This ensures full documents.
reproducibility and auditability.

High Throughput Cost-effective Storage

Nomic Embed provides high-quality, compact Nomic Embed offers flexible embedding sizes
embeddings, ideal for high-throughput, via Matryoshka representation learning. Users
data-heavy workflows. On an AWS Sagemaker can choose to store 64, 128, 256, or 512
single GPU ml.g5.xlarge instance, it returns an embedding dimensions from the full 768.
embedding roughly every 0.01 seconds. Smaller sizes reduce performance loss and
storage costs linearly.

Unleashing Nomic Embeddings with MongoDB Atlas

● Seamless Integration: MongoDB Atlas ● Streaming and Triggers: Mongo Stream
integrates Nomic embeddings effortlessly, Processing is a perfect fit for Nomic Embed’s
storing both embeddings and metadata in high throughput capabilities. Incoming data
MongoDB collections, either together or streams are robustly processed and can be
separately. Both MongoDB Atlas and Nomic combined with MongoDB Triggers to
Embed are available on AWS Marketplace for generate embeddings for immediate
identical VPC deployments. downstream use. Given Nomic Embed’s
lightweight nature and offline capabilities
● Powerful Analytics Capabilities: MongoDB
(via private or local deployments from open
Vector Search combines Nomic embeddings
source), embeddings can be produced and
for fast semantic search, enabling the fusion
ingested into MongoDB at extremely rapid
of vector search and traditional database
transfer rates.
queries on metadata. It's a flexible analytics
tool for data insights, user recommendations,
and more.

MAAP Partner
 MongoDB Atlas

Figure 76: Use-case: PDF Search

PDF Search is a use-case that combines the ● Large PDFs can be chunked and
capabilities of Nomic Embed and MongoDB ingested into MongoDB Atlas via
Atlas for an accessible, high-throughput stream processing, while
solution. Nomic Embed simplifies the ● Nomic Embed can quickly produce
process of embedding PDF previews directly long-context embeddings from the
into your application, while MongoDB Atlas processed text.
provides a powerful and scalable NoSQL ● MongoDB Vector Search integrates
database to store and index your PDFs for semantic search on Nomic
efficient searching. This combination allows embeddings with traditional database
you to quickly build a user-friendly search queries for multi-faceted downstream
experience for your PDFs without worrying analysis.
about complex infrastructure management.

© 2024 MongoDB, Inc. All rights reserved.

Together AI: Building Super-Fast
GenAI Apps with Inference,
Embeddings, and MongoDB Atlas

Founded in San Francisco in 2022, Together AI is the AI Acceleration Cloud for building and running
generative AI (gen AI). Over 150,000 developers, and organizations like Zomato, The Washington
Post and DuckDuckGo run gen AI in production using Together’s Platform. The company has raised
over $200 million, counting NVIDIA, Salesforce Ventures, Kleiner Perkins, Lux, and NEA as investors.

Together Inference Together Embeddings and Rerank APIs

Run super-fast inference at production scale for Build Retrieval-Augmented Generation (RAG)
models like Llama-3 or your own custom models, applications with ease, with access to leading
with up to 4x faster performance than vLLM and open-source embeddings and rerank models
2x faster than hyperscalers. Deploy effortlessly through Together's easy-to-use APIs, at up to
with our developer-friendly APIs and scale easily. 12x lower cost than proprietary solutions.

“We prioritized integrating with MongoDB because of its relevance and importance in the
AI stack.”
Vipul Ved Prakash
Founder and CEO at Together AI

Build Better AI Apps Faster

By integrating MongoDB Atlas with Together ● Reduced Complexity and Cost MongoDB
AI’s inference and embedding capabilities, Atlas and Together AI simplify RAG app
developers can build AI applications that development by making it easy to keep
deliver fast, real-time insights. This joint solution embeddings up to date. Together AI also
enables AI models to leverage RAG to provides high-performance inference at
recommend accurate, data-driven results that significantly lower costs than closed-source
meet user criteria while maintaining accuracy. solutions.

● Faster Time-to-Market By integrating

MongoDB Atlas with Together APIs,
developers achieve up to 4x faster inference
for models like Llama-3, accelerating
development and speeding time-to-market.

MAAP Partner
Figure 77: Together AI framework

Unlock the Power of Open-Source Embeddings for 12x Less

The Together Embeddings endpoint offers To demonstrate this integration, the

access to eight leading open-source engineering team at Together AI created a
embedding models at up to 12x cheaper tutorial for developers exploring how to build
price than proprietary alternatives. The list of a RAG application with MongoDB Atlas. This
the models includes top models from the tutorial shows how to use Together
MTEB leaderboard (Massive Text Embedding Embeddings and Together Inference to
Benchmark), such as UAE-Large-v1 and BGE generate embeddings and language
models, and state-of-the-art long context responses.
retrieval models. Together Embeddings also
offers integrations to MongoDB Atlas,
LangChain, and LlamaIndex for RAG.

© 2024 MongoDB, Inc. All rights reserved.

Own your GenAI with Arcee AI

Arcee AI meets you where you are on your AI journey, always giving you full ownership of your
models and data. We provide companies with cutting-edge out-of-the-box models, state-of-the-art
custom models, and/or our easy-to-use model training platform. You get the freedom to deploy the
models to any environment–including SaaS, dedicated SaaS, on-premise, or VPC–with guaranteed
data privacy and security. Our solutions cater to various use cases from customer service to
software development, offering flexible pricing options and dedicated support. Our featured
products include:
SuperNova & Other Top-Tier Models Advanced Model Refinement
High-performance general use models Our best-in-class post-training pipeline,
out-of-the-box, like SuperNova, our distilled incorporating synthetic dataset generation,
version of Llama-405B that outperforms SFT, reward modeling, sparse auto-encoding,
leading models in various benchmarks. model merging/infusion, and DPO to produce
state-of-the-art models.

Optimized AI models for edge devices Swarm: A Domain-Specific Model Network

Small custom models including A network of highly-specialized “expert” models
device-optimized options like Arcee Ember (Swarm) that outperform generalist LLMs in
(1.5B) and Arcee Pulse (3.8B) for edge devices. specific domains.

Unleashing Arcee AI with MongoDB Atlas

● Simple, Straightforward Deploymentless: ● Cost Conscious: With many customers
MongoDB Atlas users can access SuperNova working in a usage-based environment,
through the AWS Marketplace by subscribing Arcee AI offers both the 70B parameter
to the model. A SageMaker CloudFormation SuperNova and the 8B parameter SuperNova
script then configures and deploys Lite which is also extremely powerful but has
SuperNova to a SageMaker endpoint. the cost-efficiency advantages of being very
small.
● Superpowered Search: SuperNova provides
advanced insights and can answer complex,
multi-step questions. This empowers Atlas
users with deeper understanding of their
data and makes it easier and faster for them
to surface relevant information.

MAAP Partner
© 2024 MongoDB, Inc. All rights reserved.
Figure 78: MongoDB Atlas and Arcee AI integrated in AWS for real-time, AI-driven
responses using RAG.

The integration of MongoDB Atlas and Arcee ● Scalability: MongoDB Atlas ensures
AI, as depicted in the diagram, enables flexible, cloud-native scaling to handle
efficient data retrieval and AI-driven growing data needs without
responses through a seamless pipeline. compromising performance.
MongoDB's scalable cloud infrastructure ● AI-Augmented Retrieval: Arcee AI
ensures fast access to data using RAG
enhances retrieved data with
indexes, allowing Arcee AI to augment the
advanced generative processing for
retrieval with powerful generative
more accurate and meaningful
capabilities. This combination enhances the
speed and accuracy of responses, making it responses.
ideal for real-time applications like ● Real-Time Performance: The
interactive AI systems. MongoDB's reliability combined architecture delivers fast,
and Arcee AI's advanced inference reliable query handling suitable for
capabilities complement each other, interactive and time-sensitive
resulting in a unified model response. applications.
Overall, this architecture enables dynamic,
scalable, and intelligent query handling in
cloud environments.

© 2024 MongoDB, Inc. All rights reserved.

How GoBots AI for E-commerce
Increases Retailer Sales
Conversion by 40%

Major retail brands have long been using various forms of AI, for example statistical analysis and
machine learning models, to better serve their customers. But with its high barriers to entry, one key
channel has been slower to embrace the technology. By connecting large and small brands with
customers, e-commerce marketplaces such as Amazon, Mercado Libre, and Shopify are among the
fastest growing retail routes to market. Since 2016, GoBots has been working to extend the benefits
of AI to any retailer on any marketplace. It uses AI, analytics, and MongoDB Atlas to make
e-commerce easier, more convenient, and smarter for brands serving Latin America.

GoBots increases engagement and conversion rates for over 600 clients across Latin America,
including Adidas, Bosch, Canon, Chevrolet, Dell, Electrolux, Hering, HP, Nike, and Samsung.

The solution makes the benefits of AI available to any retailer, whether large or small. With the
GoBots natural language understanding (NLU) model, retailers automate customer interactions such
as answering questions and resolving issues through intelligent assistants. At the same time, they
leverage data analytics to offer personalized customer experiences.

By using GoBots AI for ecommerce with MongoDB Atlas, customers have grown sales conversions by
40% and reduced time to customer response by 72%.

With the power of MongoDB’s developer data platform and flexibility of MongoDB’s
document model, GoBots builds higher-performing AI-powered applications faster:

● MongoDB Atlas provides a single data platform enrich historical questions with outputs
that serves multiple operational and AI use cases. generated by different models and compare the
This includes user data and product catalogs as results. This means that they are not blocked
well as a store for AI model inferences, outputs of behind complex schema changes that would
multiple AI models for experimentation and otherwise slow down the pace of harnessing new
evaluation purposes, a data source for fine-tuning data in their models for training and inference.
models, and for vector search.
● The question-answer pairs output by the
● GoBots is evaluating the use of Atlas Triggers for company’s NLU models and LLMs are complex
invoking AI model API calls in an event-driven data structures with many nested entities and
manner as the underlying data changes. arrays. Being able to persist these directly to the
database without first having to transform them
● The flexibility provided by MongoDB’s document
into a tabular structure improves developer
model allows the development team to continually
productivity and reduces application latency.
Figure 79: GoBots question processing architecture

GoBots’ custom NLU models are built using question-answer generation with all
the Rasa framework with a neural network inferences also stored in MongoDB. If the
trained on over 150 million question-answer models are able to generate an answer with
examples and more than 50 bots — high confidence, the GoBots service will
specialists in different segments — to respond directly to the customer in real time.
understand more specific questions. In case of a low confidence response, the
models flag the question to a customer
Models are fine tuned with data from the service representative who receives a
retailer's own product catalog and website pre-generated suggested response.
corpus. The model runtime is powered by a
PyTorch microservice on Google Cloud. The With all question-answer pairs from the
larger GoBots platform is built with Kotlin different models written to the MongoDB
and orchestrated by Kubernetes, providing Atlas database, the data is used to further
the company with cloud freedom as its tune the natural language models while also
business expands and evolves. guiding model evaluations. The company has
also recently started using Atlas Vector
The GoBots AI assistants kick into action as Search to identify and retrieve semantically
soon as a customer asks a question on the similar answers to past questions. The search
marketplace site, with the questions stored results power a co-pilot-like experience for
in MongoDB Atlas. GoBots’ natural language customer service representatives and
models are programmatically called via a provide in-context training to its fleet of
REST API to perform tasks like named entity LLMs.
recognition (NER), user intent detection, and

© 2024 MongoDB, Inc. All rights reserved.

Story Tools Studio Brings Gen AI
To Gaming With Myth Maker AI

Story Tools Studio harnesses cutting-edge generative AI (gen AI) technologies to craft immersive,
personalized, and infinite storytelling experiences. Their flagship game Myth Maker AI leverages
MUSE (Modular User Story Engine), an internally developed AI-powered, expert-guided story
generator that blends a growing collection of advanced AI technology with creative artistry to weave
real time narratives.

MUSE (Modular User Story Engine) combines professionally crafted stories with user-empowered
experiences. Players make intentional choices that guide the story with AI adapting to each decision
in real time, providing a unique and personalized journey. MUSE separates the story from game
mechanics, allowing the development of multiple game types. Its use of AI creates more agile teams
with fewer dependencies.

“By selecting MongoDB, we were able to create a prototype of our game in just 48 hours. It is only
with MongoDB that we can release new features to production multiple times per day. We
couldnʼt achieve any of this with a relational database.”

Roy Altman
Founder and CEO at Story Tools Studio

AI, transactions, and analytics with MongoDB

The engineering team has used MongoDB Atlas By running MongoDB in Atlas, Story Tools
from the very start of the company. MongoDB Studio’s engineering team:
stores all of the data used in the platform:
● Is free to focus on AI-driven gaming
— user data, scripts, characters, worlds, coins,
experiences, and not on the grind of
and prompts are all richly structured objects
managing a database
stored natively in MongoDB. The games are
● Was able to scale seamlessly and
built in React and Javascript.
automatically as the team graduated from
Beyond gameplay, the company’s developers their closed beta into public beta
are now exploring MongoDB’s ACID ● Manage player demands with dozens of
transactional integrity to support in-game new players every day – every 24 hours they
monetization, alongside in-app intelligence to are organically adding dozens of new players
further improve the gaming experience through with tens of gigabytes of new data streaming
player analytics. into the platform
Figure 80: Story Tools modular user experience engine

MUSE orchestrates multimodel gen AI to create real time, unlimited stories

When a player starts a game in Myth Maker the team is working on generating enhanced
AI, they are presented with the option to 3D assets and video from text prompts. With
choose their starting hero character. Under the pace of AI advancement, the creativity
the covers, MUSE calls the GPT4 API, which of the team, and the input from game
takes the player’s selection and writes a fully testers, Story Tools Studio has the flexibility
customized adventure premise. From that to continuously deploy new features with
initial personalized script, MUSE MongoDB’s dynamic and flexible document
programmatically calls specialized AI models data model. This enables Story Tools Studio
to collaboratively generate an immersive, to build a truly innovative, artistic platform,
multimodal gaming experience using opening up a whole new world of
images, animation, audio, and soon, video experiences for both creators and audiences
and 3D. alike.

For story generation and text to voice, this is

run in Azure’s OpenAI service. Visual assets
are created via Leonardo AI, and the team
are constantly experimenting with new
models to create richer modalities. Currently,

© 2024 MongoDB, Inc. All rights reserved.

Accelerating App Development
With the Codeium AI Toolkit

Of the many use cases set to be transformed by generative AI (gen AI), the bleeding edge of this
revolution is underway with software development. Developers are using gen AI to improve
productivity by writing higher-quality code faster. Tasks include autocompleting code, writing docs,
generating tests, and answering natural language queries across a code base. How does this translate
to adoption? A recent survey showed 44% of new code being committed was written by an AI code
assistant.

Codeium is one of the leaders in the fast-growing AI code assistant space. Its AI toolkit is used by
hundreds of thousands of developers for more than 70 languages across more than 40 IDEs including
Visual Studio Code, the JetBrains suite, Eclipse, and Jupyter Notebooks. The company describes its
toolkit as “the modern coding superpower,” reflected by its recent $65 million Series B funding round
and five-star reviews across extension marketplaces. Codeium was developed by a team of
researchers and engineers to build on the industry-wide momentum around large language models,
specifically for code. They realized that their specialized generative models, when deployed on their
world-class optimized deep learning serving software, could provide users with top-quality AI-based
products at the lowest possible costs.

Training models on MongoDB

Codeium has recently trained its models on “MongoDB is wildly popular across the
MongoDB code, libraries, and documentation. developer community. This is because
Now developers building apps with MongoDB
Atlas integrates the fully managed
can install the Codeium extension on the IDE of
their choice and enjoy rapid code completion database services that provide a unified
and codebase-aware chat and search. developer experience across
Developers can stay in the flow while they build, transactional, analytical, and generative
coding at the speed of thought, knowing that AI apps.”
Codeium has ingested MongoDB best practices
Anshul Ramachandran
and documentation.
Head of Enterprise & Partnerships at Codeium
Getting Started with MongoDB and Codeium

MongoDB APIs are incredibly powerful, but In its announcement blog post and YouTube
due to the breadth and richness of the APIs, video, the Codeium team shows how to build
it is possible for developers to be spending an app in VSCode with MongoDB serving as
more time than necessary looking through the data layer. Developers can ask questions
API documentation or using the APIs on how to read and write to the database,
inefficiently for the task at hand. An AI get code competition suggestions, explore
assistant, if trained properly, can effectively specific functions and syntax, handle errors,
assist the developer in retrieval and usage and more. This was all done at no cost using
quality of these APIs. Unlike other AI code the MongoDB Atlas free tier and Codeium
assistants, we at Codeium build our LLMs 100% free.
from scratch and own the underlying data
layer. This means we accelerate and optimize You can get started today by registering for
the developer experience in unique and novel MongoDB Atlas and then downloading the
ways unmatched by others. Codeium extension.

© 2024 MongoDB, Inc. All rights reserved.

 Putting Jina AIʼs Breakthrough
Open Source Embedding Model
To Work

Jina AI has swiftly risen as a leader in multimodal AI, focusing on prompt engineering and
embedding models. With its commitment to open-source and open research, Jina AI is bridging the
gap between advanced AI theory and the real world AI-powered applications being built by
developers and data scientists. Over 400,000 users are registered to use the Jina AI platform.

Jina AI’s work in embedding models has caught significant industry interest. As many developers
now know, embeddings are essential to generative AI (gen AI). Embedding models are sophisticated
algorithms that transform and embed data of any structure into multi-dimensional numerical
encodings called vectors. These vectors give data semantic meaning by capturing its patterns and
relationships. This means we can analyze and search for unstructured data in the same way we’ve
always been able to with structured business data. Considering that over 80% of the data we create
every day is unstructured, we start to appreciate how transformational embeddings — when
combined with a powerful solution such as MongoDB Atlas Vector Search — are for gen AI.

“Our Embedding API is natively integrated with key technologies within the gen AI
developer stack including MongoDB Atlas, LangChain, LlamaIndex, Dify, and
Haystack. MongoDB Atlas unifies application data and vector embeddings in a
single platform, keeping both fully synced. Atlas Triggers keeps embeddings fresh
by calling our Embeddings API whenever data is inserted or updated in the
database. This integrated approach makes developers more productive as they
build new, cutting-edge AI-powered apps for the business.”
Dr. Han Xiao
Founder and CEO at Jina AI
Figure 81: Jina AI’s world-class embedding models improve search and RAG systems.

Jina AI’s embedding models

Jina AI's jina-embeddings-v2 is the first In our published Jina Embeddings v2 and
open-source 8K text embedding model. Its MongoDB Atlas article we show developers
8K token length provides deeper context how to get started in bringing vector
comprehension, significantly enhancing embeddings into their apps. The article
accuracy and relevance for tasks like covers:
retrieval-augmented generation (RAG) and
semantic search. Jina AI’s embeddings offer 1. Creating a MongoDB Atlas instance
enhanced data indexing and search and loading it with your data. (The
capabilities, along with bilingual support. The article uses a sample Airbnb reviews
embedding models are focused on singular data set.)
languages and language pairs, ensuring 2. Creating embeddings for the data
state-of-the-art performance on set using the Jina Embeddings API.
language-specific benchmarks. Currently, 3. Storing and indexing the
Jina Embeddings v2 includes bilingual embeddings with Atlas Vector
German-English and Chinese-English models, Search.
with other bilingual models in the works. 4. Implementing semantic search using
the embeddings.
Jina AI’s embedding models excel in
classification, reranking, retrieval, and
summarization, making them suitable for
diverse applications, especially those that
are cross-lingual. Recent examples from
multinational enterprise customers include
the automation of sales sequences, skills
matching in HR applications, and payment
reconciliation with fraud detection.

© 2024 MongoDB, Inc. All rights reserved.

Superduper: Build Next-Gen AI
Apps on Your Existing Database

Superduper.io, provides an open-source framework for implementing AI applications and workflows

with their existing databases. By transforming the database into a central AI platform and
consolidating all critical steps and components of AI applications within a single environment, AI
teams can implement custom AI solutions without the need for data migration via pipelines or
complex MLOps infrastructure. From GenAI, vector search, and RAG to classic machine learning.

Their enterprise platform is built for efficient and secure scalability, self-hosted on existing data
infrastructure (whether in the cloud or on-prem), so that data does not need to be shared externally.

Superduper offers different AI app and workflow templates ready-to-install on MongoDB and
highly configurable with minimal development effort. Because the application templates are based
on their open-source development framework, you can fully own the codebases and control any level
of implementation detail. Current use-cases include Multi-Modal Vector Search & RAG, Document
Extraction & Analysis, Anomaly Detection, Visual Object Detection, and Image and Video Search.

Find an executive summary here.

“We integrate MongoDB as a core backend database for our platform, using the
PyMongo driver for app connectivity and Atlas Vector Search for vector
embedding storage and querying. Having MongoDB Ventures on board enables us
to work even more closely with their teams, optimizing our product while
engaging with MongoDBʼs vast developer community.”

Duncan Blythe
Co-Founder & CTO of Superduper.io
Figure 82: AI application lifecycle with Superduper.io

Development Features Deployment Features

● Integrations with major AI frameworks, ● Compute running where data resides,

model hubs, and APIs and compatible minimizing data movement on existing
with any Python functionality infrastructure, whether cloud or on-prem

● Ability to switch between and combine ● Scalable model hosting, low latency
different tools and providers, even in the streaming inference and model
same workflow training/fine-tuning

● Operating and navigating data simply ● Database transformation into an AI

with database queries (instead of platform, including feature store and model
building pipelines) hub

● A simple Python interface with reusable ● Granular management and observability

building blocks user interface

● Application templates for different AI ● Relying on existing security and rights

use-cases, ready-to-install on the management controls of the datastore
database

© 2024 MongoDB, Inc. All rights reserved.

4149.AI: Maximizing team
productivity with a hypertasking
AI-powered teammate

4149.AI boosts team productivity with a dedicated AI teammate. In a successful private beta, nearly
1,000 teams leveraged this agent to streamline goal tracking and tasks. It analyzes team
communication, identifies roadblocks, and takes action in Slack discussions, meetings, calls, reports,
emails, and task trackers.

AI-powered team Participation in tasks

4149.AI provides teams with their own The AI agent participates in various tasks such
AI-powered teammate that helps track goals as joining slack threads, meetings, transcribing
and priorities. calls, generating summaries, responding to
emails, and updating issue trackers.

No-code customization Ambitious Growth Strategy

There is a no-code way for people to customize 4149.AI outlines an aggressive roadmap for its
and expand the functionality of their AI products, leveraging the power of
teammate. chain-of-thought reasoning and multimodal
capabilities in advanced language models.

The Power of Unified Data

● The ability to store summaries and chat ● Being part of MongoDB’s AI Innovators
history alongside vector embeddings in program provides 4149.AI with access to
the same database accelerates developer technical support and free Atlas credits,
velocity and the release of new features. helping them quickly experiment using the
native AI capabilities available in the
● The hybrid search capability of MongoDB
MongoDB developer data platform.
Atlas allows pre-filtering data with
keyword-based Atlas Search before
semantically searching vectors, which helps
retrieve relevant information faster.
4149.AI helps teams get more work done by The engineers at 4149.AI evaluated multiple
providing them with their very own vector stores before deciding on Atlas Vector
AI-powered teammate. During the Search. The ability to store summaries and
company’s private beta program, the chat history alongside vector embeddings in
autonomous AI agent has been used by the same database accelerates developer
close to 1,000 teams to help them track velocity and the release of new features. It
goals and priorities. It does this by building also simplifies the technology stack by
an understanding of team dynamics and eliminating unnecessary data movement.
unblocking key tasks. It participates in slack
threads, joins meetings, transcribes calls, Looking forward 4149.AI has an aggressive
generates summaries from reports and roadmap for its products as it starts to more
whitepapers, responds to emails, updates fully exploit the chain-of-thought and
issue trackers, and more. multimodal capabilities provided by the
most advanced language models. This will
4149.AI uses a custom-built AI-agent enable the AI co-worker to handle more
framework leveraging a combination of creative tasks requiring deep reasoning such
embedding models and LLMs from OpenAI as conducting market research, monitoring
and AI21 Labs, with text generation and the competitive landscape, and helping
entity extraction managed by Langchain. identify new candidates for job vacancies.
The models process project documentation The goal for these AI teammates is for them
and team interactions, persisting summaries to eventually be able to take the initiative in
and associated vector embeddings into what to do next rather than rely on someone
Atlas Vector Search. There is even a no-code to manually assign them a task.
way for people to customize and expand the
functionality of their AI teammate. Over
time, the accumulated context generated for
each team means more and more tasks can
be offloaded to their AI-powered co-worker.

© 2024 MongoDB, Inc. All rights reserved.

Zelta.AI: Prioritizing product
roadmaps with data-driven
customer analytics

In the rapidly evolving digital economy, Zelta.AI stands as a beacon for product managers
navigating the sea of customer feedback. Born out of the need to synthesize diverse feedback into
coherent development plans, Zelta.AI is revolutionizing the way businesses prioritize their product
roadmaps

Generative AI for Customer Insights Processing Unstructured Data

Zelta uses generative AI to communicate Zelta leverages Language Models (LLMs) to
insights on top of customer pain points found in process unstructured data and returns
companies’ most valuable asset: qualitative actionable insights for product teams.
sources of customer feedback such as call
transcripts and tickets.

Integration with Multiple Platforms Real-Time Product Feedback Trends

Zelta.AI has the capability to pull data directly Zelta.AI offers real-time product feedback
from multiple platforms like Gong, Zoom, trend reporting, enabling faster decisions for
Fireflies, Zendesk, Jira, Intercom, among product teams, enhancing its value.
others.

Choice of MongoDB as Developer Data Platform

● MongoDB provides Zelta with the flexibility to provide customers with real-time
to constantly experiment with new features. dashboards and reporting of trends in
They can add fields and evolve the data product feedback.
model as needed without any of the
● Looking forward, as Zelta plans on creating
expensive schema migration pains imposed
its own custom models, MongoDB will prove
by relational databases.
invaluable as a source of labeled data for
● Zelta makes heavy use of the MongoDB supervised model training.
aggregation pipeline for application-driven
intelligence. Without having to ETL data out
of MongoDB, they can analyze data in place
Figure 83: Zelta leverages LLMs to process unstructured data and returns actionable insights for
product teams

The company’s engineering team uses a “We also make heavy use of the MongoDB
combination of fine-tuned OpenAI GPT-4, aggregation pipeline for application-driven
Cohere, and Anthropic models to extract, intelligence. Without having to ETL data out
classify, and encode source data into trends of MongoDB, we can analyze data in place
and sentiment around specific topics and to provide customers with real-time
features. MongoDB Atlas is used as the data dashboards and reporting of trends in
storage layer for source metadata and product feedback. This helps them make
model outputs. product decisions faster, making our service
more valuable to them.”
“The flexibility MongoDB provides us has
been unbelievable. My development team Mick Cunningham
can constantly experiment with new CTO and Co-Founder at Zelta AI
features, just adding fields and evolving the
data model as needed without any of the
expensive schema migration pains imposed
by relational databases.”

Mick Cunningham
CTO and Co-Founder at Zelta AI

© 2024 MongoDB, Inc. All rights reserved.

Crewmate: Helping brands
connect with their communities

Crewmate is a no-code builder for embedded AI-powered communities. The company’s builder
provides customizable communities for brands to deploy directly onto their websites. Crewmate is
already used today across companies in consumer packaged goods (CPG), B2B SaaS, gaming,
Web3, and more.

Customizable Communities Context-Aware Semantic Search

Crewmate creates AI-powered communities for Crewmate's platform features context-aware
SaaS firms, boosting sales, retention, and search powered by Atlas Vector Search,
engagement. Users can interact, share insights, delivering relevant content to users on brand
and discuss your product. community pages.

Real-Time Data Pipelines Insightful Analytics for Brands

Crewmate implements event-driven pipelines to Crewmate enables brands to extract valuable
ensure that community content remains fresh insights from user engagement data using its
and up-to-date. powerful analytics capabilities.

Choice of MongoDB as Developer Data Platform

● MongoDB Atlas provides integrations with stored in MongoDB, without the need to
the fast-evolving AI ecosystem. Crewmate extract, transform, and load (ETL) it into a
leverages this capability to easily integrate separate data warehouse or data lake.
with other AI models, such as OpenAI's
● Crewmate utilizes Atlas Vector Search, a
ada-002 and potentially other models like
feature provided by MongoDB Atlas, to
Llama in the future.
power context-aware semantic search.
● Crewmate utilizes MongoDB's Query API to This enables users visiting a brand's website
process, aggregate, and analyze user to automatically access relevant content
engagement data. This allows brands to such as social media posts, forum
track community outreach efforts and discussions, job postings, and special offers.
conversions directly from the app data
Personalized Community Content with Atlas Vector Search

Using context-aware semantic search Thaker goes on to say, “The introduction of

powered by Atlas Vector Search, users Atlas Vector Search and the Building
hitting and browsing the community pages
generative AI Applications tutorial gave me
on a brand’s website are automatically
a fast, ready-made blueprint that brings
served relevant content. This includes posts
from social media feeds, forum discussions, together a database for source data, vector
job postings, special offers, and more. search for AI-powered semantic search,
and reactive, real-time data pipelines to
“Iʼve used MongoDB in past projects and keep everything updated, all in a single
knew that its flexible document schema platform with a single copy of the data and
would allow me to store data of any
a unified developer API. This keeps my
structure. This is particularly important
engineering team productive and my tech
when ingesting many different types of
stack streamlined. Atlas also provides
data from my clientsʼ websites,”
integrations with the fast-evolving AI
Raj Thaker ecosystem. So while today Iʼm using
CTO and Co-Founder of Crewmate OpenAI models, I have the flexibility to
easily integrate with other models, such as
Llama, in the future.”

© 2024 MongoDB, Inc. All rights reserved.

Video personalization at scale
with Potion and MongoDB

Potion enables salespeople to personalize prospecting videos at scale. Already over 7,500 sales
professionals at companies including SAP, AppsFlyer, CaptivateIQ, and Opensense are using
SendPotion to increase response rates, book more meetings, and build customer trust.

Effortless Video Creation Efficient Outreach

Sales representatives simply record a video With Potion, you can engage, convert, and
template and select the elements they want to leave a lasting impact on your prospects. It
personalize. These elements typically include eliminates the need for manual video recording,
details like the recipient’s name, company, and streamlining your communication efforts.
desired call-to-action.

Bulk Transformation AI Models and Technologies

Imagine turning a single video template into Potion’s custom generative AI models are built
over 1,000 unique video messages, each using PyTorch and TensorFlow. Their vision
tailored to an individual contact. Potion model is trained on thousands of faces,
achieves this by efficiently reanimating videos allowing them to synthesize videos without
in bulk, saving time and effort for sales teams. individualized AI training. Audio models are
tuned on-demand for each voice.

“We use the MongoDB database to store metadata for all the videos, including the
source content for personalization, such as the contact list and calls to action. For
every new contact entry created in MongoDB, a video is generated for it using our
AI models, and a link to that video is stored back in the database. MongoDB also
powers all of our application analytics and intelligence. With the insights we
generate from MongoDB, we can see how users interact with the service,
capturing feedback loops, response rates, video watchtimes, and more. This data
is used to continuously train and tune our models in Sagemaker."
Kanad Bahalkar
Co-Founder & CEO at Potion
Scaling Potion with MongoDB Atlas

On selecting MongoDB Kanad says, “I had To further enrich the SendPotion service,
prior experience of MongoDB and knew Kanad is planning to use more of the
developer features within MongoDB Atlas.
how easy and fast it was to get started for
This includes Atlas Vector Search to power
both modeling and querying the data. Atlas
AI-driven semantic search and RAG for
provides the best-managed database users who are exploring recommendations
experience out there, meaning we can across video libraries. The engineering team
safely offload running the database to is also planning on using Atlas Triggers to
MongoDB. This ease-of-use, speed, and enable event-driven processing of new
video content.
efficiency are all critical as we build and
scale the business." Potion is a member of the MongoDB AI
Innovators program. Asked about the value
of the program, Kanad responds, “Access to
free credits helped support rapid build and
experimentation on top of MongoDB,
coupled with access to technical guidance
and support."

© 2024 MongoDB, Inc. All rights reserved.

Artificial Nerds: The power of
custom voice bots without the
complexity of fine-tuning
Artificial Nerds, founded in 2017, is a software company that unlocks the potential of AI for
businesses through a suite of intelligent virtual assistants. Their custom voice bots streamline
customer interactions, allowing teams to focus on building meaningful relationships.

Human-Like Conversations Voicebot Integration

Artificial Nerds’ AI bots are designed to create Artificial Nerds offers voicebots that automate
fluid and personalized conversations with phone calls for businesses. These voicebots
customers. Unlike traditional bots with scripted understand user language and provide
responses, Artificial Nerds’ innovative tools solutions, making business interactions more
ensure a more natural and user-centric efficient and effective.
experience.

No-Code Builder Contact Center Enhancement

Their platform enables easy adjustments to Artificial Nerds streamlines tasks, boosting
chatbots without coding. Using cloud tech and efficiency and freeing up teams for strategic
templates, businesses can act quickly, cutting work. Their advanced bots offer visibility,
development time. control, and real-time adjustments, all without
developer intervention.

Enabling Flexible Data Storage for AI-Powered Products

● Artificial Nerds chose MongoDB for its ● By eliminating the need for a separate
flexible schema, which allows them to store search engine and ETL, MongoDB Atlas
richly structured conversation history, reduces the complexity of development and
messages, and user data. This flexibility is management. This allows developers to focus
crucial for a company focused on on building their application without worrying
AI-powered products, as it enables them to about maintaining separate data stores.
adapt and evolve their data structures as
needed to support their evolving suite of
products and services.
AI-Fueled Search and Innovation: Artificial Nerds Speeds Up with MongoDB Atlas

By adopting Atlas Search, the company Artificial Nerds is growing fast, with revenues
streamlined its search capabilities, expanding 8% every month. The company
integrating a powerful full-text index directly continues to push the boundaries of
onto its database collections. This eliminated customer service by experimenting with new
the need for separate search engines and models including the Llama 2 LLM and
ETL mechanisms, reducing cognitive multilingual sentence transformers hosted in
overhead. Similarly, the release of Atlas Hugging Face. Being part of the MongoDB
Vector Search further enhanced efficiency by AI Innovators program helps Artificial Nerds
replacing a standalone vector database with stay abreast of all of the latest MongoDB
MongoDB Atlas, resulting in improved product enhancements and provides the
developer productivity and a 4x reduction in company with free Atlas credits to build new
latency for a better customer experience. features.

© 2024 MongoDB, Inc. All rights reserved.

Algomo: Conversational support,
powered by generative AI

Algomo uses generative AI to help companies offer their best service to both their customers and
employees across more than 100 languages. The company’s name is a portmanteau of the words
Algorithm (originating from Arabic) and Homo, (human in Latin). It reflects the two core design
principles underlying Algomo’s products:

AI Agents with Human-Level Reasoning Personalized Interactions

Algomo provides AI agents with human-like Algomo's AI chatbot delivers personalized
understanding and decision-making interactions, tailoring content to individual
capabilities, enhancing customer service by customers, posing clarifying questions, and
efficiently managing tasks and seamlessly capable of communicating in over 100+
transitioning complex issues to support teams. languages.

Efficient Help Desk Omnichannel Support

Algomo's Helpdesk integrates teams, channels, Algomo's Messenger offers customization
and data into a single workspace, simplifying options to align with any brand and enables the
support operations. Furthermore, Algomo's AI reception of messages from multiple channels,
operates in Co-Pilot Mode, offering suggestions such as email, WhatsApp, and social media.
to enhance the efficiency of customer service
teams.

Alamo Optimizes Support with MongoDB Atlas

● Alamo chose MongoDB due to its flexible ● Alamo's engineers are considering Atlas
document data model, allowing them to Vector Search as a replacement for their
store customer data alongside conversation current standalone vector database. This
history and messages, ensuring long-term move not only reduces costs but also
memory for context and continuity in support simplifies their codebase by eliminating the
interactions. need to synchronize data across two
separate systems.
● MongoDB Atlas as a fully managed cloud
service relieves Alamo's team from
operational heavy lifting, enabling them to
focus on building conversational
experiences rather than managing
infrastructure.
Figure 84: Algomo Bot

Unlocking Personalized Customer Experiences with Algomo's Conversational AI

With Algomo, customers can get a Customers can instantly turn their support
ChatGPT-powered bot up on their site in less articles, past conversations, slack channels,
than 3 minutes. More than just a bot, Algomo Notion pages, Google Docs, and content on
also provides a complete conversational their public website into personalized
platform. This includes Question-Answering answers. Algomo vectorizes customer
text generators and autonomous agents that content, using that alongside OpenAI’s
triage and orchestrate support processes, ChatGPT. The company uses RAG (Retrieval
escalating to human support staff for live Augmented Generation) prompting to inject
chat as needed. It works across any relevant context to LLM prompts and
communication channel from web and Chain-Of-Thought prompting to increase
Google Chat to Intercom, Slack, WhatsApp, answer accuracy. A fine-tuned
and more. implementation of BERT is also used to
classify user intent and retrieve custom
FAQs.

© 2024 MongoDB, Inc. All rights reserved.

Conclusion

Across industries, AI has captured the

imaginations of executives and consumers
alike. Whether you’re a customer of a bank,
insurance company, telecommunications
enterprise, or retail conglomerate, AI has and
will transform and enhance the way you do
business with corporations. For the industries
that matter most globally, AI has created
opportunities to minimize risk and fraud,
perfect user experiences, and save
companies from wasting labor and resources.

MongoDB Atlas will revolutionize industries’ To learn more about industry-specific

abilities to incorporate operational, solutions for AI developers, visit the
analytical, and generative AI data services. MongoDB Solutions Library to access
Leading companies like Bosch and Telefonica reference architectures, product guides,
use MongoDB in their AI-enhanced IoT and key tools for building your next
platforms, while Iguazio uses MongoDB as generative AI application. If you are ready
the persistence layer for its data science and to dive in even further with our experts,
MLOps platform. schedule an Innovation Workshop with our
team today.
From creation to launch, MongoDB Atlas
guarantees that AI applications are
cemented in accurate operational data and
fulfill the demands of scalability, security,
and performance by developers and
consumers alike.
 Next Steps with
MongoDB
MongoDB's unique blend of speed, flexibility, and robust security offers a
compelling proposition for organizations building AI-enriched applications. Our
ability to provide a scalable, resilient, and efficient data management solution,
deployment flexibility and support for multi-cloud strategies positions MongoDB as
a leader for intelligent applications.

The conversation about leveraging MongoDB within industries doesn't end here. We
invite you to delve deeper into MongoDB's capabilities and offerings to discover
how you can build the future of AI applications.

Contact us at [email protected]

Innovation workshops AI resources

Learn about out Industries program to Get full access to our resources to Build
take advantage of innovation AI-powered Apps including articles,
workshops and more. reports, case studies and more.

© 2024 MongoDB, Inc. All rights reserved.