Local_Pdf_Chat_RAG

Why we included this project

For developers who want to see how retrieval-augmented generation actually works instead of treating a framework as a black box, this is a readable reference implementation. The pipeline is split into modules for document parsing, chunking, embedding with FAISS, BM25 keyword retrieval, hybrid search, reranking, and answer generation, and each piece can be read, tested, or swapped on its own. Both a Gradio interface and a FastAPI endpoint expose the same workflow, so you can upload a PDF and ask questions locally without much setup. Ollama and OpenAI-compatible backends are supported, which keeps offline experimentation practical. The repo is upfront that this is a teaching tool, not a hardened service, and it flags the limitations you should expect before trusting it with real data.