Scaffold-ETH Docs & SRE Challenges assistant

Open source AI assistant based in RAG (Retrieval Augmented Generation), to help users resolve their SpeedRunEthereum questions, and let users "chat" with Scaffold-ETH docs.

Tech stack

Generic RAG Diagram

This assistant MVP uses LangChain framework with these providers:

• GroqCloud: to interact with LLM via API. We can easily plug different Models like (check GroqCloud docs for specific model versions):
  • llama-3.3-70b-versatile (current model)
  • qwen
  • deekseek
  • gpt oss
  • gemma
• Google: to create the embeddings
• FAISS: for vector store and to search for embeddings that are similar to user prompt

Everything is wrapped in Streamlit to transform the Python script into a web app.

Knowledge base

Each Challenge will have it's own set of documents in a [Challenge #] folder. It will be loaded when the user selects the Challenge in the dropdown menu. By default, [Challenge 0] docs are loaded.

Document list for each challenge:

• Scaffold ETH docs
• Challenge readme
• Telegram Q/A extracted from Challenge chat group