A RAG (Retrieval-Augmented Generation) system that turns PDF documents into a queryable knowledge base — answers grounded strictly in your content, not AI training data. Built with FastAPI, ChromaDB, and Google Gemini. PDFs are chunked into 500-word segments, embedded via gemini-embedding-2, and stored as vectors. Gemini LLM answers using only the top-3 retrieved chunks, eliminating hallucination entirely. Key features: - Auto model fallback chain across Gemini variants for free-tier reliability - Full REST API (upload, query, delete) via FastAPI - CLI mode for direct terminal queries - Dual-interface design for flexibility