A question-answering system that gives source-backed answers from PDF, DOCX and text files. It combines meaning-based (vector) search with keyword (BM25) search, and a query classifier routes each question to the best search strategy. Overlapping chunks keep context across page breaks, and every answer records which passages it used, so results can be traced and checked. It's split into ingestion, retrieval and generation layers, with a knowledge base manager and a usage dashboard.
FastAPI · Gemini · BM25 · SQLite