RAG Chatbot Over 200 PDFs — Python, LangChain, Pinecone, OpenAI

We're a small consulting firm with about 200 PDFs built up over the years: client deliverables, market research, internal SOPs, case studies. Every week someone loses an hour digging for an answer that's already documented. I want an internal knowledge base chatbot that answers from those documents and cites its sources. Required experience: Has shipped production RAG (retrieval-augmented generation) systems, not prototypes or tutorial projects Strong Python, with FastAPI for backend services LangChain for the retrieval pipeline (LlamaIndex experience also relevant) Pinecone, or a comparable vector database such as pgvector, Weaviate, or Qdrant OpenAI API, GPT-4o or similar, for answer generation Document ingestion at scale: PDF parsing, chunking strategy, embeddings, and metadata for source attribution Semantic search and vector search over a multi-document corpus Supabase for authentication and per-user chat history Next.js, React, and TypeScript, enough to build a clean internal chat interface Prior work on document Q&A, knowledge base chatbots, or LLM integration with citations What gets built: a web chat UI where the team asks a question and gets an answer grounded only in our documents, returning the source PDF name and page number with every response. No hallucination from the model's general knowledge. If the documents don't cover it, the bot says so rather than guessing. Engagement: core retrieval pipeline within a week, then tuning once we see performance on our real documents. Internal use only. Likely phase 2 for Slack integration and role-based access per team. Suitable for an AI engineer, LLM developer, RAG engineer, or full-stack Python developer with genuine retrieval experience. Skills: Python · LangChain · Retrieval Augmented Generation · Pinecone · OpenAI API Vector Database · Chatbot Development · Generative AI · Natural Language Processing · LLM Prompt Engineering