Professional Summary
GenAI Engineer & Data Scientist with 4 years of experience building and shipping production LLM, RAG, and agentic-AI systems in enterprise settings. Owned end-to-end delivery of a production RAG finance assistant, an org-wide AI-native SDLC framework, and an agentic ServiceNow bot. Deep hands-on Azure AI stack (AI Search, OpenAI, Functions), multi-agent orchestration (Semantic Kernel, AutoGen), and Graph-RAG — turning GenAI into measurable business outcomes across finance, supply chain, and engineering.
Technical Skills
Professional Experience
★ Flagship GenAI Projects
RAG Finance Application
Production RAG assistant over 1–2k internal finance documents (reports, contracts, compliance) for natural-language Q&A and search. Built the full Python pipeline — ingestion → unstructured chunking → embeddings → retrieval — on Azure AI Search + Azure OpenAI.
AI-Native SDLC Framework
Org-wide brownfield AI-development framework: builds a reusable context layer (domain, code-mapping, patterns, guardrails, hooks), enforces TDD, runs multi-agent code review, and auto-creates PRs — powered by GitHub Copilot.
ServiceNow Agentic Bot
An agentic IT-support system on Semantic Kernel + Azure OpenAI with a vector-indexed KB and SQL incident store. Interprets tickets, auto-resolves routine ones (password resets, access), escalates the rest.
Additional GenAI & ML Projects
- CodeWiki Bot (Graph-RAG): converted an entire codebase into a Graph-RAG knowledge graph via AST parsing (Tree-sitter); powers a chatbot, automatic documentation, architecture-diagram generation, and a bug finder. Standalone, end-to-end.
- SRE Agents (next-gen reliability): configured & deployed Microsoft's Azure SRE-Agent for specific teams — autonomous incident triage with broad connector support — as the follow-on to the ServiceNow bot. MTTR ↓ ~80%. Standalone, end-to-end.
- Legacy Code Converter: a sequential swarm of AutoGen agents (Azure OpenAI) converting ~400 legacy
.scriptfiles to PySpark with auto-docs — replacing ~1 month of manual work. Standalone, end-to-end. - AI Test-Case Generator: GenAI tool generating maximal test cases directly from BRDs — saving 50–60 hrs of manual test-case writing per cycle. Standalone, end-to-end.
- Synthetic Data + Phi-3 SLM Fine-tuning (Finance): a two-stage privacy-first pipeline — generated 100,000+ synthetic financial records at 95% statistical similarity using NVIDIA Nemotron-4 340B on Azure Durable Functions, then fine-tuned Microsoft Phi-3-mini (LoRA on Azure ML) on that data into a private, domain-expert finance-Q&A SLM — chosen over a large LLM for privacy, lower cost, and deep domain knowledge. Deployed to Azure AI Foundry.
- Supply-Chain Insights Chatbot (POC): Azure-OpenAI GPT chatbot for business-data insights with metadata-driven query generation on a Python/FastAPI RAG backend — a client-facing proof of concept.
Earlier Data-Science Engagements
- B2B Lead Generation (global hospitality chain): analyzed 11M+ small/mid-business records; ran EDA and engineered features for a lead-classification model.
- Digital Transformation — Middle East central bank: built interactive Qlik Sense dashboards, optimized data scripts, and integrated data sources.
Education
NIT Silchar · CGPA 8.1/10
Board of Secondary Education, Assam · 86%