naitik-2006/web3_rag_pipeline — reverse-engineered prompt

Build me a Python RAG assistant for Web3 and Bitcoin development research.

I want it to have two modes. One mode should work like a general research chatbot that can pull useful info from Arxiv and web search, rewrite the user’s question, retrieve relevant context, and answer without making things up. The other mode should answer questions about the Bitcoin dev mailing list or another codebase repo by using a saved FAISS index plus a graph of files, functions, classes, and relationships so it can reason across connected pieces.

Please set up the ingestion scripts, indexing, prompts, environment config, and simple command line ways to run both modes. Use Together and Groq APIs for embeddings and fast Llama responses, with API keys coming from a .env file. Include a way to extract emails into JSONL and rebuild the static index when the mailing list or repo changes.

Keep it modular and easy to swap in another repo later. Look up current docs online if you need to.