Mangesh W.

I build RAG systems and AI agents for document-heavy applications, from source retrieval and tool workflows to the web interface. My work includes a 500-paper (nearly 8 million tokens) research assistant built with Python, FastAPI and LangGraph. I help teams build document assistants, add AI workflows to existing products, and investigate retrieval, integration and backend reliability issues. RELEVANT WORK • Research retrieval and GraphRAG Built query routing, hybrid retrieval, reranking, and retrieval of text, tables and figures. For questions spanning multiple papers, I used Neo4j and Leiden communities to discover connected evidence, with original source passages supporting cited answers. • Document-aware AI agents Built textbook retrieval organized by book, chapter, concept and page. The voice tutor uses application tools to fetch relevant text and diagram context and open the corresponding PDF page while explaining it. • Caching and reliable execution Implemented shared Valkey caching, versioned request identities and duplicate-request coordination for the research assistant. PostgreSQL-backed jobs, bounded workers and replayable events support long-running workflows and reconnecting clients. • Full-stack SaaS development Implemented and maintained subscriptions, memberships, payment integrations, webhooks and generation workflows in an existing SaaS application. This includes tracing failures across the interface, API, database and external providers. One client's feedback: “His handover write-ups are the clearest I have worked with.” HOW I WORK For an existing system, I start with representative questions, documents or workflow traces to identify where the problem occurs. We agree on a focused milestone and acceptance checks before expanding the scope. You get implementation updates, verification notes, and a clear account of what has been tested and what still needs configuration. Core stack: Python, FastAPI, LangGraph, Qdrant, Neo4j, PostgreSQL, Next.js/React, Docker and AWS. Send a few representative documents or questions and describe the workflow you want to support. We can define the first milestone and how to verify it.