The whole discipline, in the order you actually need it: from your first model call to running the thing at 3am when retrieval returns nothing and the agent will not stop retrying.
For software engineers who can already ship a service and now have to ship one with a model in it.
Most production AI failures are not model failures. They are a model being asked to do something it was never able to do reliably, by an engineer who had not been told where the edge was. This lesson draws that edge.
LLM Strengths & Limitations
No. This is engineering, not model training. You need to be comfortable writing and testing a backend service; the course supplies everything about the model. There is no linear algebra and nothing is trained from scratch.
Because most production systems that need AI features are already written in something else, and the engineers asked to add them are backend engineers. The concepts are language-independent; the runnable code is Java and Spring Boot so it drops into the codebase you actually work in.
Each lesson pairs a short read with something you run. The projects are downloadable and self-contained, and the failure labs are diagnosis exercises against a system that is already broken — you read the evidence and work out the cause.
Roughly 28 hours of lesson time, spread across nine modules. Engineers doing it alongside a job typically take two to three months and stop at the module their work needs.
LLM Foundations if you have never shipped an AI feature. If you have one in production and it is misbehaving, start at the module that names your problem — retrieval, agents, memory or evaluation — and come back for the rest.