At ThoriumTek, I developed a LangChain and Pinecone RAG assistant that gives sales and support teams cited answers from product datasheets and service documents. I evaluated retrieval approaches with DeepEval and LangSmith, improving retrieval quality by 35% and reducing hallucinations by 28%.
At Spandan Society, I built LLM workflows to classify and route volunteer and donor requests, generate donor impact reports with human approval, and match volunteers to events. These systems used tools including LangGraph, FastAPI, PostgreSQL, Slack, and Sheets.
In my projects, I built a payment recovery orchestration engine and an enterprise gateway for controlling AI-initiated database access. I also developed a multi-agent researcher with a critique-and-revise loop and structured tool access through MCP.
