rag-demystified

Why we included this project

Anyone who has leaned on LlamaIndex or Haystack and wondered what really happens between their query and the final answer will get a lot out of this project. It reimplements the advanced machinery of retrieval-augmented generation in plain Python on top of EvaDB, with a runnable example that answers complex questions over a set of Wikipedia articles about cities. The author's central point is that much of that machinery, from deciding which sub-questions to ask and which data source each should target to choosing between vector and summary retrieval, reduces to carefully designed LLM prompt templates rather than hidden framework magic. Along the way the repo covers the trade-offs and failure points of advanced RAG, exactly the transparency developers want before they trust a framework's abstractions in production. It is a concise, runnable reference for teams evaluating RAG tooling or anyone teaching themselves how sub-question query engines work.