Conversational Document
Retrieval System
Making large enterprise document repositories easier to explore through grounded, conversational search.
Explore case studyFull Stack Gen AI Engineer / Building Scalable AI Solutions / Kolkata, West Bengal, India /
Making AI feel
like good engineering.
I build LLM-powered products that connect useful ideas to dependable systems—from conversational document retrieval to agents that make business data easier to reach.
Retrieval, reasoning, and the details that make them useful.
01 — A working modelInputs to insight
01A real question
02Relevant context
03A useful answer
Built withcare & context
01 The practice
Curious by nature.
Practical by design.
Full Stack Generative AI Engineer specializing in scalable AI systems and enterprise-grade RAG architectures.
Designed and delivered reliable multi-LLM orchestration pipelines, conversational AI systems, and intelligent document processing solutions. Strong expertise in backend engineering (Django, FastAPI), vector search, and production optimization. Passionate about building reliable, high-impact AI applications.
More about my experience02 Selected work
Systems with a
point of view.
A selection of AI engineering work, described through the problems, architecture, and implementation behind each system.
Conversational Document
Natural Language
to SQL Agent
Turning natural-language questions into schema-aware SQL with generation, validation, and refinement.
Explore case studyAutomated Quality
Assurance Agent
Helping quality teams review lab-report parameters against regulatory thresholds and spot discrepancies.
Explore case studyEarlier work Academic projects
A couple of earlier projects developed alongside my studies.
Blog App Backend Django
It's a Django-based Blog App where users can create, publish, and manage posts. Features include a user-friendly admin panel, tagging system, RSS feed, search, email sharing, commenting, and tag-based post recommendations.
Associated with Maulana Abul Kalam Azad University of Technology, West Bengal (formerly WBUT)
Tools Django, django-taggit, psycopg2, markdown
Wikipedia Scrapper
This project is built to scrape a Wikipedia page based on the topic provided by the user and display a summarized version of the topic along with all the reference and image links. It will help the user to learn about the topic in a shorter amount of time.
Associated with Maulana Abul Kalam Azad University of Technology, West Bengal (formerly WBUT)
Tools Selenium, Flask, JSON, NLTK, ChromeDriverManager, etc.
03 Behind the systems
The thinking behind
the build.
Each project starts with a practical problem. Here’s how the pieces fit together.
01
Retrieval · Conversation · Context
Conversational Document
Retrieval System
The problem
Business users needed a more accessible way to find information across enterprise document repositories.
The architecture
Qdrant vector retrieval paired with embedding models via Azure OpenAI and Vertex AI, coordinated through a multi-step LangGraph pipeline with conversational memory.
The implementation
Designed chunking and retrieval, built Django/DRF backend APIs, and developed a responsive chat interface for real-time querying.
The outcome
Reduced document-search effort and improved information accessibility for business users; no quantitative result was specified.
02
Orchestration · Validation · Data access
Natural Language
to SQL Agent
The problem
Make relational business data easier to query without requiring every user to write SQL manually.
The architecture
A multi-LLM workflow using Azure OpenAI and Gemini through Vertex AI for query generation, validation, and refinement.
The implementation
Added schema-aware prompting and structured output parsing, then connected query execution to a results dashboard with conversational follow-ups.
The outcome
Accelerated business data access and reduced dependence on manual SQL querying; no quantitative result was specified.
03
Document intelligence · Rules · Reporting
Automated Quality
Assurance Agent
The problem
Quality teams needed to check lab-report values against FSSAI regulatory thresholds and identify discrepancies.
The architecture
An AI-driven document-intelligence workflow combining LLM-based structured parameter extraction with rule-based validation.
The implementation
Built parameter extraction, threshold comparison, discrepancy detection, and reporting workflows for QA teams and vendors.
The outcome
Reduced manual compliance-verification effort while improving accuracy and turnaround time; no numeric measurements were provided.
04 Experience
Good work is
built over time.
A path through full-stack development, applied AI, and systems designed to make everyday work a little easier.
Aug 2024 — Present
Full-time · Hybrid
Full-time · Hybrid
IT Executive — ISA · Full Stack Gen AI Engineer
ITC Infotech · Kolkata, West Bengal, India
Designing and deploying scalable Generative AI systems for enterprise clients, including conversational RAG architectures, multi-LLM NL-to-SQL pipelines, and AI-driven document intelligence agents. Responsible for backend system design, vector search integration, conversational memory implementation, and production optimization using Azure OpenAI and Vertex AI.
Jul 2023 — Jul 2024
Full-time · On-site
Full-time · On-site
Junior Python Developer
AM Infoweb Pvt. Ltd. · Pune District, Maharashtra, India
- Designed and developed a custom medical records portal for a major US healthcare client, including requirements gathering, backend infrastructure, and frontend prototype development. Managed the project lifecycle independently and used Agile methodology for iterative development and timely milestone delivery.
- Managed the company’s internal Django-based product, handling requirements gathering, backend modifications, and occasional frontend adjustments.
- Spearheaded development and implementation of a Redaction feature, enhancing employee productivity and product performance. Also contributed to LLM-assisted medical-record document workflows, including chronological organization and automated delivery.
- Implemented Selenium-based Robotic Process Automation (RPA) solutions to reduce manual effort and optimize operations.
- Collaborated with stakeholders to integrate new functionalities, prioritizing user experience and system efficiency.
- Contributed to code reviews and continuous improvement of development processes within the team.
05 Technical expertise
Tools for the
whole journey.
From a model call to the application around it, I enjoy bringing the full system into focus.
Generative AI & retrieval
Generative AI, Retrieval-Augmented Generation (RAG), Agentic AI Development, LangChain, LangGraph, Google ADK, Azure OpenAI, Vertex AI, HuggingFace, OpenRouter, Ollama, LiteLLM, Qdrant, BM25, ColBERT, ColPali, hybrid retrieval, embeddings, metadata filtering, MCP, conversational memory, HITL workflows, guardrails, structured outputs, response evaluation.
Backend & data
Python, Django, Django REST Framework, FastAPI, REST APIs, PostgreSQL, MySQL, MongoDB, Redis, Apache Kafka, asynchronous processing, multiprocessing, multithreading, schema design, query optimization, unit testing.
Frontend & product interfaces
HTML, CSS, JavaScript, Bootstrap, jQuery, Streamlit, responsive interfaces, AJAX/fetch, dynamic rendering, data visualization integration, pagination, lazy loading, Selenium UI automation.
Platforms & delivery
Microsoft Azure OpenAI, Google Cloud Platform, Vertex AI, Docker, Git, GitHub, GitLab, service-account-based cloud integration.
06 Learning & foundations
Grounded in
curiosity.
Education
Maulana Abul Kalam Azad University of Technology, West Bengal (formerly WBUT)
Oct 2020 – Jun 2022 · 1st Class
University of Calcutta
Jun 2016 – Jun 2019 · 1st Class
Score: 103 · Oct 2020 · AIR 29
Courses & certifications
The next good question
Let’s make
something useful.
Interested in thoughtful GenAI engineering and practical product work? I’d be glad to connect.
✉ Contact me