kallappa2005/docu-rag ? reverse-engineered prompt

Build me a FastAPI app that lets users upload PDF files and then ask questions about them in a chat style interface.

I want the app to extract text from the PDFs, split it into useful chunks, turn it into embeddings, and store it in a persistent Chroma database so the documents can be searched later. When someone asks a question, it should find the most relevant parts of the uploaded files and use a Groq powered language model to generate a helpful answer with conversation memory so follow up questions still make sense.

Please include a simple health check endpoint, Swagger docs, clean request and response models, and good logging. Use environment based settings for things like API keys, chunk size, top results, upload folder, and database path. Make it production style and easy to run locally with uv. If you need to check current library docs or best practices, look them up online first.