anthonysandesh/QA-RAG-SQL ? reverse-engineered prompt

Build me a chatbot app that lets people ask questions in plain English about tabular data.

I want it to work with a read only SQLite database, plus CSV and XLSX files. It should let me chat with a preloaded dataset, and also upload new CSV or XLSX files during the session, process them, then switch back to chatting with them. The answers should come from the data when possible, using retrieval augmented generation for better context, and it should handle simple SQL style questions safely without ever allowing writes or destructive actions.

Use GPT 3.5 with LangChain, and store or search the file based data with a vector database like Chroma. I also want a simple web UI where I can choose whether I am chatting with SQL data, processed tabular files, or uploaded files. If you need to check current docs for anything, feel free to look them up online.