Project — Document Q&A Chatbot with RAG

Build a fast, accurate chatbot that lets users query a collection of documents in natural language and receive reliable answers with citations linking back to the source. The system should: Ingest documents in multiple formats (PDF, Word, HTML), with semantic chunking optimized for retrieval accuracy. Use hybrid retrieval — vector similarity search combined with keyword search (BM25), followed by a re-ranking step to surface the most relevant chunks first. Generate grounded answers via LLM, using only the retrieved text as context. Every answer includes inline citations to the specific source document and section. Refuse to answer when the documents are silent on the topic. No hallucination, no guesses — clear "this is not covered in the provided documents" responses. Be fast — target less than 3 seconds end-to-end for typical queries. Streaming responses for perceived speed. Caching layer for repeated questions. Provide a simple web interface — chat-style UI where users ask questions, see streamed answers with citations, and can click citations to view source content.