I've built production-grade GenAI systems including V-RAG, a verified multi-agent RAG pipeline for evidence-grounded legal question answering. Using LangChain, LangGraph, LangSmith, OpenAI, Chroma, and Pydantic, I designed Responder, Prosecutor, and Judge agents that detect unsupported claims and revise answers.
My evaluation work includes a 50-question suite covering retrieval, temporal reasoning, and contradictions, reaching 0.745 case-level recall and 0.951 MRR at k=7. I also built a document-compliance agent that combines OCR, PDF parsing, PostgreSQL, LangChain, and Gemini to automate regulatory review across structured and unstructured documents.
I've independently delivered full-stack products for Aseries Envirotek, Digiproctor, Trajectorie, and freelance clients, using React.js, Node.js, PostgreSQL, and JavaScript. I also built HOPE, a GPT and Whisper-powered CBT platform that reached 60+ monthly active users and received Microsoft for Startups backing.
