Engineers and technical product people who already use LLM APIs and want the model behind them to stop being a black box. It suits anyone building retrieval or agent systems who keeps hitting behaviour they cannot explain, developers moving into an AI-facing role, and people preparing for GenAI interviews where the questions go one layer below the API. No machine-learning background is assumed — the course teaches transformer intuition rather than the linear algebra — but you should be comfortable reasoning about how a system behaves.

Each lesson opens with a short set of reading cards on the concepts it covers, then puts you straight into questions. Every question is multiple choice and every wrong answer produces an explanation immediately, so a mistake becomes the thing you remember rather than something you look up later. Questions in this course start at medium difficulty — the trivial recall items from the full course are left out, because a crash course should be spent on the ideas that actually decide an interview.

Your accuracy is tracked per lesson, so after a single pass you have a map of which of the twelve areas you are solid on and which need another round.

A video course is measured in hours watched; this is measured in questions answered. Watching someone explain temperature feels productive, but you find out whether you understood it the moment you have to choose between four plausible answers about what changes when you raise it. Recall beats recognition, and questions force recall. The practical difference is speed: 380 targeted questions take a few hours and tell you exactly where you are weak, where a twenty-hour video series leaves you feeling informed but untested.

The trade-off is honest — a video series is better when a topic is entirely new to you. That is why this course assumes you have seen LLM foundations before, and why the full course exists for the topics where you want more room.

The difference is depth, not coverage. Both reach the same twelve areas. This one gives you one merged lesson per area cluster and 380 questions; GenAI Engineering: LLM Foundations breaks the same material into 172 lessons with roughly ten questions each, so you can sit on a single idea until it is genuinely automatic. Start here if you are short on time or want to find your gaps; move to the full course for the areas the gaps turn up in. Once the foundations are solid, RAG Systems Crash Course and AI Agents Crash Course build directly on them.

Speed comes from cutting repetition, not from cutting material. A conventional syllabus teaches temperature, top-k and top-p as three separate lessons, each reopening the same setup about sampling from a probability distribution over tokens. Merged into one lesson, that setup is stated once and the three controls land together — which is also how you actually tune them, jointly rather than one at a time.

The second lever is difficulty. Roughly a quarter of the questions in a full syllabus are recall checks — what does this acronym stand for, which company released that model. They matter for orientation and they are noise when you are building. Those are left out, so every question here is about behaviour rather than trivia.

A practical order: work straight through the twelve areas without stopping to perfect anything, note the lessons where your accuracy drops below about seventy percent, then do a second pass on only those. Tokenization and evaluation are the two areas most people are weakest on, and both have direct cost and quality consequences in production.