// the find
alexeygrigorev/llm-rag-workshop
Chat with your own data - LLM+RAG workshop
A workshop repo teaching RAG fundamentals: index FAQ docs into Elasticsearch, retrieve relevant chunks, pass them to an LLM. Aimed at developers who want a working RAG pipeline in an afternoon without fighting infrastructure. Companion material to the DataTalks LLM Zoomcamp course.
The progression from raw retrieval to generation to a Streamlit UI is well-paced — you build something real at each step before moving on. Supporting both OpenAI and Groq with identical code, plus Ollama for offline use, means you can follow along regardless of what you have access to. The evaluation section covering hit rate, MRR, and cosine similarity between generated and original answers is more rigorous than most workshop repos bother with. Docker Compose setup for Ollama + Elasticsearch removes the 'works on my machine' friction.
The README is the entire workshop — there are no standalone scripts you can run end-to-end; you have to reconstruct the logic from fragments scattered across the document. `rag.py` is referenced but its content is never shown, so anyone returning to the repo after the live session loses the finished state. Elasticsearch is hardcoded to `localhost:9200` everywhere with no config layer, which breaks the moment you deploy anywhere. The evaluation section is tacked on as an 'alternative' with external links rather than included code, so the most useful part of the workshop is the hardest to actually run.