National Conference on
Engineering Innovations in Emerging
Technologies (NCEIET-2025)
DEPARTMENT OF INFORMATION SCIENCE AND ENGINEERING
Approved by AICTE, New Delhi | Affiliated to VTU, Belagavi | Accredited by NBA
Jnana Prabha, Bidarahalli, Virgo Nagar Post, Bengaluru, Karnataka 560049
Date: 23rd & 24th May 2025
Autonomous AI Agents for Task
Automation and Advanced Reasoning
• Guide:
INDUMATHI
Asst.Professor
Department of ISE,EPCET
Authors:
SAQULAIN ULLA SHARIFF N (1EP22IS096)
NAVEEN S B (1EP22IS071)
PAVANKALAYAN K V (1EP22IS074)
PAVAN KUMAR B (1EP22IS075)
Abstract
:
• Autonomous AI agents perform tasks without human intervention by combining
perception,planning, and reasoning. Using large language models (LLMs) and decision-making
algorithms, they automate complex work flows in fields like health, business, finance, and software
development.
• Frameworks such as Auto GPT and Lang Chain enable these agents to interface with tools,
remember information, and adaptto task changes. Advanced reasoning techniques like chain-of-
thought prompting enhance problem-solving capabilities.
Introduction to AI Agents
and
Challenge
AI Agent Evolution
s
Agent-based LLMs like GPT and BERT have transformed natural language
processing tasks such as chatting, translation, and summarization.
Applications
These agents automate complex tasks in healthcare, finance, education, and customer
service, improving productivity by handling mundane tasks like meetingtranscriptionand
communication assistance.
AI Agent Definition and
Features
• What is an AI Agent?
• AI agents are intelligent virtual assistants that autonomously perceive, analyze, decide, and act to achieve
goals. Unlike traditional tools, they are proactive, learn from feedback, and adapt over time.
• Benefits
• Automate repetitive tasks like email drafting and scheduling Analyze large datasets to free humans for
strategic work Collaborate with other agents for problem-solving
Example
• Tanz, as ales person, uses her AI assist anttodraft client emails, suggest follow-ups, and automate
scheduling, allowing her to focus on high-value activities.
Literature Review: Frameworks and
Architectures
LLM and Reinforcement Learning
Combining LLMs with reinforcement learning enables agents to learn behaviors by
rewarding desired actions, as shown in robotics applications.
Embodied Agents
Agents respond to verbal commands using multimodal transformers processing vision
and language inputs.
Multi-Agent Communication
LLMs facilitate einter-agent communication to enhance coordination and problem-
solving.
Challenges
High computational demands and interpretability remain issues, prompting research into
hybrid and adaptive models.
Comparative Analysis of
LLM Architecture
s
1. Eneric LLMs
Offer broadcapabilities butmaylack depth inspecialized domains.
2. Domain –Specific Models
Improve performance by incorporating relevant knowledge and
context.
3. Task Optimization
Custom models canbe optimized for speed, accuracy, and compliance with
regulations. Eneric LLMs.3.Task Optimization
Applications of LLM-Based Autonomous
Agents
Social Science
Agents simulate psychology experiments, support mental health, and assist research by
generating abstracts and study scripts.
Research and Experiment Assistance
Agents automate planning, design, and execution of scientific experiments, using tools like Chem
Crow to support chemical studies safely and effectively.
System Architecture
and
Workflow of AI Agent
Step 1: Creating a Strategy
User and AI agent collaborate to define goals and create a customized plan, such as assessing autonomous
agent projects.
Step 2: Choosing Instruments
The agent selects the best tools based on compatibility with the strategy, like APIs or data analysis tools.
Step3: Implementing Actions
The agent activates tools to execute tasks, analyze data, rank results, and refine its approach.
Example: Hotel Reservation
The AI books ahotelmatchingp references, manages reservations, arranges airport pickup, and sends
reminders and notifications.
Problems Encountered
and
Prototype
Demonstration
Role-Playing Capability
Agents must adopt specific roles like researcher or programmer, but LLMs struggle with novel roles and
lack human-like comprehension.
Solutions
Fine-tuning with human data and developing new AI architectures can improve performance, though
building these systems is challenging.
Expected Outcomes and
Conclusion
The rise of AI agents using LLMs like GPT-4 marks a revolutionary era in software engineering,
improving corporate procedures and software development.
debugging time, better coding standards, and conflict resolution.and creativity.
The proposed AI agent architecture supports continuous learning, task execution, and smooth
integration into complex environments.
Ongoing research will expand applications across industries, transforming operations, communication
and creativity,
Autonomous AI Agent: Hotel Booking
Workflow
1.Detects Context
• Checks calendar for trip: Business to NYC, May 12–15, 2025. Notes preferences: Near Central Park, gym, free breakfast,
≤$200/night.
1.Autonomous Search and Decision
• Searches booking sites (Booking.com,Expedia).Filters for hotels near Central Park with gym and
breakfast.Compares ratings and prices. Picks Parkside Inn: $189/night, 4.5 stars, 0.4 miles away.
