A containerized RAG system that answers questions over a mixed-document corpus (PDFs, docs, text) and returns the exact source files behind every answer. Built with Python/FastAPI, a vector index, and an open-source LLM served on AMD GPUs with ROCm.