GomuGomuNo01/Chatbot-RAG

DocAssist is a Retrieval-Augmented Generation assistant to query an organization’s documents and get sourced answers (file, page, excerpt). It’s meant for teams that need fast, precise lookup across contracts, manuals, policies, HR files, etc. Key features: - Sourced answers with file/page/excerpt and explicit refusal when answer isn’t in the corpus. - Multi-format ingestion: PDF (≤500 pages), .docx, .txt, .md; chunking (~900 characters). - Workspaces to separate domains (HR, legal, technical). - Hybrid retrieval: semantic vectors (FAISS) + lexical BM25 with dynamic weighted RRF. - Cross-encoder reranking (BGE) and Claude Haiku LLM to generate answers from top passages only. - Semantic cache, conversational memory, daily usage limits, incremental indexing (MD5), persistent storage on Cloudflare R2 and FAISS indices on HuggingFace Hub. - Built with FastAPI, LangChain orchestration, BAAI bge-small embeddings, PyMuPDF/python-docx parsing.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.