A production-style RAG system — point it at any folder of documents and ask questions in plain English; it answers only from your documents and cites every claim by source number.

• Hybrid retrieval fusing FAISS (semantic) and BM25 (keyword) search via Reciprocal Rank Fusion, then a cross-encoder reranker re-scores the top candidates against the question.
• Provider-agnostic LLM layer (Gemini, Mistral, OpenAI, Anthropic — swap via one .env line) with a citation-enforcing prompt, so every answer is traceable to source text.
• 23 automated tests (12 embedder unit tests, 11 pipeline integration tests).