llm-applications

Why we included this project

This one comes from the Ray team, so it treats a RAG pipeline as something that has to hold up under load rather than a notebook you run once. It walks the whole stack, from loading and chunking documents through embedding, indexing, and serving, and it covers the parts tutorials usually skip: measuring retrieval quality per component and against the final answer, so you can compare configurations before committing. There is also a worked example of routing queries between open models and hosted APIs like OpenAI when cost or quality favors one over the other. Teams past the prototype stage will find a concrete reference for the scaling decisions, not just the happy path.