profile-pic

Utsav Dey

I am a Data Scientist and Agentic AI Engineer with experience delivering enterprise-grade AI and Generative AI solutions in regulated and client-driven environments. My background spans advanced analytics, machine learning, NLP, and LLM-based architectures, with a strong focus on building scalable, secure, and explainable AI systems aligned with business and compliance requirements.

In my current role, I work on Agentic AI and Generative AI initiatives, designing end-to-end LLM pipelines using Model Context Protocol (MCP), Retrieval-Augmented Generation (RAG), and tool-based orchestration. I have developed Reason-Action agent frameworks that enable natural language interaction with complex enterprise APIs, implemented centralized LLM Gateways, and integrated guardrails, authentication, and monitoring to ensure responsible AI usage. My work includes deploying production systems on AWS using Docker, ECS, ECR, Cognito, DynamoDB, PostgreSQL (PGVector), and FastAPI.

Previously, as a Data Scientist in the banking domain, I led large-scale NLP and analytics projects focused on customer experience, risk reduction, and operational efficiency. I analyzed over 200,000 customer complaints using supervised and transformer-based models (SVM, XGBoost, BERT), applied topic modeling techniques (LDA, BERTopic), and built RAG-based GPT solutions using open-source LLMs for domain-specific financial insights. These initiatives reduced turnaround time, improved classification accuracy, and supported data-driven decision-making at scale.

I bring strong expertise in Python, SQL, Machine Learning, Deep Learning, NLP, Transformers, Generative AI, Agentic AI, Prompt Engineering, and cloud-native AI deployments. I am particularly interested in consulting-driven problem solving—working with stakeholders, translating business requirements into technical solutions, and delivering AI systems that are robust, auditable, and aligned with enterprise standards.

I am keen to contribute to consulting engagements involving Generative AI, Advanced Analytics, Intelligent Automation, and Responsible AI

  • Role

    Agentic AI Engineer

  • Years of Experience

    5.3 years

Skillsets

  • rag
  • Machine Learning
  • Matplotlib
  • Model context protocol
  • MS-SQL
  • Natural Language Processing
  • NumPy
  • pandas
  • PGVector
  • PL-SQL
  • PostgreSQL
  • Postman
  • Prompt Engineering
  • Logistic Regression
  • Scikit-learn
  • Seaborn
  • SQL
  • Statistical Modelling
  • support vector machine
  • TensorFlow
  • Transformers
  • Unsupervised Learning
  • XgBoost
  • Llm based pipeline
  • Vector DB
  • Fastmcp
  • Agentic AI
  • AWS Amplify
  • AWS Cognito
  • AWS ECR
  • AWS ECS
  • BERT
  • context management
  • Deep Learning
  • Docker
  • DynamoDB
  • Excel
  • FastAPI
  • Python
  • Generative AI
  • Github
  • Hive
  • Java
  • Keras
  • LangChain
  • Large Language Models
  • linear algebra
  • LLAMA
  • LLM
  • Llm gateway

Professional Summary

5.3Years
  • Jul, 2025 - Jul, 2025

    Agentic AI Engineer

    Cognizant
  • May, 2023 - Jul, 20252 yr 2 months

    Data Scientist

    Bandhan Bank
  • Jul, 2019 - Dec, 20212 yr 5 months

    System Engineer

    Tata Consultancy Services

Work History

5.3Years

Agentic AI Engineer

Cognizant
Jul, 2025 - Jul, 2025
    Worked as an Agentic AI Developer for a US-based client (Lenders Toolkit) on a POC project. Developed an Agentic AI solution featuring a Pipeline Agent that generates payloads using various canonical names from the Encompass Mortgage Loan API. Enabled end users to retrieve loan data and details based on natural language queries. Implemented Reason and Action Agent to orchestrate tasks using MCP for tool invocation. Used LLM Gateway for invoking LLMs hosted on AWS. Applied chunking and embedding techniques for RAG functionality, with embeddings stored in AWS PostgreSQL using PGVector. Created endpoints through FastAPI and FastMCP. Developed contracts and an adapter-based approach for MCP to ensure scalability and plug-and-play functionality. Integrated tools: RAG for canonical name retrieval and reranking, AWS Guardrail for query/response monitoring and prompt attack prevention, Payload Creation Tool for dynamic payload generation via LLM, Query API for result retrieval. Secured Agent and MCP tools through Authentication service using Cognito token validation. Implemented LLM Gateway as the central hub for managing LLMs, embeddings, and Rerank models. Deployment via Docker images using AWS ECR and ECS services and frontend via AWS Amplify. Authentication managed through AWS Cognito. Used DynamoDB for memory and AWS PostgreSQL with PGVector for embeddings. Integrated CI pipeline using GitHub.

Data Scientist

Bandhan Bank
May, 2023 - Jul, 20252 yr 2 months
    Spearheaded development of a RAG-based GPT model using open source Llama model, Langchain, and VectorDB trained on bank annual reports. Enhanced model's ability to provide accurate, data-driven financial insights using advanced NLP and generative AI. Analyzed 200,000 customer complaints with classification techniques and NLP; manual annotation enabled training of models like SVM, XGBoost, and BERT. Improved classification and corrected 17 percent of misclassified complaints. Streamlined complaint resolution process, reducing time from 7 days to 48 hours. Used topic modeling (LDA, BERTopic) to uncover recurring themes for policy improvements. Applied NLP and NER to cluster non-individual Current Account customers for targeted marketing. Conducted topic modeling on app store reviews to identify product issues and enable fixes. Designed transformer classification model using BERT and NLP to filter low-quality call summaries. Increased customer engagement in campaigns by creating a propensity response model using XGBoost and Logistic Regression with 88 percent accuracy. Assessed marketing campaign effectiveness using Levene's test and independent T-test, showing significant impact.

System Engineer

Tata Consultancy Services
Jul, 2019 - Dec, 20212 yr 5 months
    Developed and implemented software solutions for a government project, ensuring adherence to client specifications. Collaborated with cross-functional teams to gather requirements, design solutions, and conduct testing, ensuring high-quality deliverables. Provided ongoing support for deployed modules, incorporating client feedback to enhance functionality. Engaged with clients to present project updates and gather feedback, fostering strong client relationships. Trained junior team members on best practices in software development. Successfully delivered multiple software modules, improving operational efficiency.

Major Projects

3Projects

Predicting Agricultural Land Boundaries After a Natural Disaster

    Predicted agricultural land boundaries in north-eastern India after natural disasters to lower insurance premium rates and settle land boundary disputes using image processing and computer vision.

EDA on FMCG Sector Stock Prices

    Performed exploratory data analysis on FMCG sector stock prices for HUL, ITC, and Colgate Palmolive, correlating closing price data with NIFTY and NIFTY FMCG index using Pearson correlation over 6 months of historical data.

Online Exam Cheating Detection System

    Developed an online exam cheating detection system using neural networks and computer vision models like RESNET.

Education

  • Post Graduation Program in Data Science

    Praxis Business School (2023)

  • B.Tech in Computer Science and Engineering

    Future Institute Of Technology (2019)