There is a growing pile of content telling you what RAG is and almost none of it walks you through building a real one and then keeping it honest in production, which is the gap LLM Zoomcamp fills. Like the rest of the DataTalks.Club catalogue it is free, open source, and unapologetically project-driven, with all the materials on GitHub and a live 2026 cohort you can register for if you want the deadlines. The seven modules are sequenced the way a practitioner actually learns this, starting with a basic RAG pipeline so you have something working almost immediately, then vector search and embeddings, then agents and tool use, and then the parts most tutorials quietly skip, evaluation, monitoring, and the best practices like hybrid search and reranking that separate a demo from a system, before finishing with an end-to-end project. I rate the course mostly for that second half.
Anyone can wire up a retrieval call against the OpenAI API, but knowing how to measure whether your retrieval is any good, how to track user feedback and system behaviour in production with something like Grafana, and how to improve results methodically is the difference between a toy and a product, and this course takes that seriously. The prerequisites are clear and fair. You need to be comfortable with Python, the command line, and Docker, but you do not need prior machine learning experience, which is the right call for a course aimed at application builders rather than researchers. That framing is also the main thing to be honest about.
This teaches you to use large language models, not to understand them. You will not learn how attention works or how these models are trained, and if that is what you are after you want Karpathy's series or the Stanford language modelling course instead. It also leans on the OpenAI API, so while the course itself is free, expect a small amount of usage cost as you work through it, and as with the other Zoomcamps the self-paced route loses the cohort deadlines that genuinely help people finish. My take is that for a working developer who wants to actually ship LLM features, this is one of the best free on-ramps available right now, current in a field that dates fast and focused on the engineering reality rather than the hype.
Treat the project as the point, build something you would be willing to show someone, and you will come out of it able to do real work.