Sohel Gafar Shaikh

Data Scientist / Machine Learning Engineer II

Email-id : [email protected]
Mobile No.: 9730595697

WORK EXPERIENCE
• Groww, Bangalore - Machine Learning Engineer II
(January 2023 - Present)
• Groww, Bangalore - Machine Learning Engineer I
(August 2020 - December 2022)
Sohel is a passionate data scientist having more than 3 years of experience across the financial services
domain. He is proficient in Python, SQL, Machine Learning and has rich experience with the entire data
science life cycle (data preparation, EDA, model development & deployment).
He has worked closely with stakeholders from multiple teams to translate various business problems into
data science use cases and has built and deployed Machine Learning solutions to solve them.

TECHNICAL SKILLS
• Languages : Python, SQL
• Frameworks : Flask, FastAPI
• Libraries : Pandas, Numpy, PyTorch, OpenCV, Scikit-Learn, Transformers
• Tools : AWS Sagemaker, AWS S3, Git, Jira, Docker, Kubernetes, Jenkins, Grafana
• ML : Structured (Tabular) data, unstructured data - Images and Text, Classification, Regression, Natural
Language Processing / NLP

PROJECTS
• Credit
◦ SMS Ledger : Created a sophisticated SMS Ledger Analysis system that extracted valuable financial
data from users’ text messages. This ledger included details on credits, debits, transaction dates, and
more, enabling the creation of personalized user profiles. By engineering features from this data, I
could provide tailored financial services, such as extracting users’ income and understanding their
spending patterns. This acts as a rich source of alternate data which will be used to improve the
existing models.
◦ Behavioural Model (B Score) : Developed the behavioral model (B Score) to assess risk for existing
loan recipients, leveraging a comprehensive blend of on-us and off-us features. This model powers
critical decisions, including repeat loan approvals, credit limit adjustments, and collection prioriti-
zation for high-risk clientele. This implementation effectively optimizes loan management processes
and enhances risk mitigation strategies.
◦ Acquisition Model (A Score) : Engineered a high-performing acquisition model through strategic
feature extraction from proprietary and credit bureau data. Effectively identified 45% of defaulters
within the top 10% of loan applicants. Deployed the model as a real-time REST API using FastAPI
framework, facilitating seamless integration and enabling efficient user underwriting. Implemented
a robust monitoring framework to ensure ongoing model performance in production.
◦ Propensity Model (P Score) : Devised a credit propensity model, forecasting users’ likelihood to
seek personal loans within two months by analyzing their platform engagement and credit bureau
data. Conducted an impactful A/B experiment, demonstrating that elevating PF and ROI for high-
propensity users enhances revenue without compromising conversion rates. This model, in conjunc-
tion with the credit acquisition model, guided dynamic pricing decisions for users’ variable interest
rates, optimizing loan offerings.
◦ Text2SQL : Led the development of a Natural Language to SQL Query Bot for BigQuery, training and
fine-tuning the T5 Model to translate plain English commands into SQL queries. This user-friendly
bot, integrated with Slack, empowered non-technical users to access BigQuery data effortlessly. De-
ployed it as a real-time API, collecting data for ongoing model improvement, significantly reducing
data retrieval time and democratizing data access within the organization.
• Growth & User Experience
◦ Revenue Prediction : Predicted on the 30th day from the date of signup, the revenue that will gained
from a user in the months 4, 5 and 6 from signup. These predictions were aggregated on the acquisition
channel level to assess the quality of users obtained from them as early as possible.
◦ Churn Prediction : Predicted the users who are going to churn from the platform and ran campaigns
over them to increase retention. Also identified the key drivers of churn so that product team can
work over them.
• Platform - Automation of the Onboarding Process
◦ Signature Verification : Built a solution that identifies whether the signature drawn by a user over a
trackpad canvas is a valid signature or not (eg a line or a dot is not a valid signature) in real time. This
reduced the time of the onboarding process from hours to minutes.
◦ Field Extraction from Documents : Classified the image uploaded by a user into following categories
and using OCR, extracted the relevant fields.
* Identity proof : Pan card
* Address proof : Aadhaar card, Driving license, Passport, Voter ID card
* Bank proof : Cancelled cheque, Passbook, Account statement
* Others : Mandate form
This solution was used to process the KYC pdf documents of more than Two Lakh users thereby
saving the bandwidth of ops agents (approx three minutes per document).
◦ Face Match : This solution matches the selfie uploaded by a user during the onboarding journey with
the photo on the address proof to identify in real time whether they are the same person or not. This
eliminated the dependency on a third party api saving the company an annual bill of more than 12
Million Rupees.

ACADEMIC DETAILS

Examination College Year Score/%

Under Graduate: BE - Computer Engineering
Graduation Pune Institute of Computer Technology, Pune 2020 9.1

HSC Yeshwant College Nanded 2016 87.8%

SSC Manovikas Vidyalaya Kandhar 2014 98.2%