Course Capstone: Graduation Checklist

Every level in the GenAI Lab is now unlocked, and every worked example in this closing tier drew on the entire course rather than introducing anything new. That is not a coincidence. It is the actual point: you now have a complete, composable toolkit, and the remaining skill is knowing how to combine it, which is exactly what the last five chapters practiced directly.

• -The difference between AI, machine learning, and deep learning, and where generative AI actually sits inside that landscape.
• -How machines learn: supervised, unsupervised, reinforcement, and self-supervised learning, the last of which is the specific mechanism that made today's LLMs possible at all.
• -The math underneath every model in this course: vectors and distance, probability, loss functions, and gradient descent.
• -Classical ML and neural networks from first principles, including why depth alone does nothing without a nonlinear activation function between layers.
• -The full data and training pipeline, evaluation discipline, and the bias-variance tradeoff that explains why a model fails and what to actually do about it.
• -AI ethics and safety fundamentals: bias, privacy, interpretability, and the discriminative-versus-generative distinction that set up everything in Phase 2.

You should be able to answer each of these out loud, in your own words, without looking anything up. If any one of them feels shaky, that is a genuinely good signal for exactly where to go back and re-read, not a reason to feel behind.

• -Why does a RAG system need chunking and reranking, and what specifically goes wrong at each stage if you skip it?
• -What is the ReAct loop, and what stops it from running forever?
• -Why is fine-tuning the wrong tool for injecting facts that change over time, and what should you reach for instead?
• -What are the two independent triggers for routing an agent's decision to a human, and why does either one alone route to a human regardless of the other?
• -Why can a system look perfectly healthy on latency, throughput, and error rate while still being badly broken?

Reading built the map. The GenAI Lab is where you actually build the territory, level by level, with real evaluation against real missions, not just multiple-choice recall. If a level ever feels harder than the reading prepared you for, that gap is the most valuable signal this entire course can give you: it is exactly where to spend the next hour, not evidence that something is wrong with you or with the course.

Congratulations on finishing AI Engineering Fundamentals. Go build something real with it.