I build RAG (Retrieval-Augmented Generation) pipelines that let your product answer questions from its own data — documents, databases, internal knowledge bases, whatever you're sitting on.
Most of my recent work has been full-stack: vector embeddings with Pinecone, chunking and retrieval logic in Python/FastAPI, and a Next.js frontend that surfaces the answers. I've built systems where users upload PDFs and get accurate, cited responses in seconds — not hallucinated summaries, actual grounded answers pulled from the right chunks.
What I typically handle: document ingestion and preprocessing, embedding pipelines, hybrid search (semantic + keyword), context window management, and response generation tuned for accuracy over fluency. I've used OpenAI, Gemini, and open-source models depending on the client's latency and cost constraints.
If you have unstructured data that your team can't query — contracts, reports, support tickets, product docs — and you want a system that can actually answer questions about it reliably, that's the kind of problem I work on.