doc-johnson/rag-chatbot — reverse-engineered prompt

Build me a local document chat app where I can upload PDF, text, or Markdown files, then ask questions and get answers based only on those documents. I want a simple web page for uploading files, choosing a model provider, chatting, seeing the answer stream in real time, and viewing the sources that were used.

Please make the backend usable on its own through a REST API, and make a basic Streamlit interface on top of it. It should support Groq and OpenAI, with Groq as an easy free option, and let me switch models from the UI. Keep chat history between sessions with SQLite.

For search, combine normal keyword matching with semantic search so it works for both exact terms and general meaning. Use local embeddings so that part doesn’t need an API key. Please don’t use LangChain or LlamaIndex, I want the retrieval logic written directly so it’s easy to understand and debug.

Add Docker setup and a clear env example. Look up current docs online if needed.