Short, self-contained lessons. Each one teaches a single idea tied to the mission: walk into a customer's problem and build a reliable solution on foundation models. Built on high-trust sources — Chip Huyen's AI Engineering, Andrej Karpathy's lectures, and Anthropic's engineering guides.

Lessons

Lesson 01 · Foundations What an LLM Actually Is The next-token-predictor mental model — and how hallucination, non-determinism, amnesia, and prompt sensitivity all fall out of it. ⏱ ~11 min lesson · 🧪 ~15 min exercise
Lesson 02 · Foundations Tokens, Context & the Cost of a Call The three numbers behind every scoping conversation — plus your first hands-on lab: a real API call, the usage meter, and time-to-first-token. ⏱ ~11 min lesson · 🧪 ~30–45 min lab
Lesson 03 · Core skills Prompt Engineering as an FDE Skill Why wording moves the distribution, the five techniques that reliably improve quality, and a lab that upgrades one prompt three times — measured, not vibed. ⏱ ~12 min lesson · 🧪 ~30 min lab
Lesson 04 · Core skills Structured Output & Function Calling Making the model programmable: schemas the API enforces, and the tool_use hand-off — the model asks, your code acts. ⏱ ~12 min lesson · 🧪 ~30 min lab
Lesson 05 · Core skills Tool Use & the Agent Loop An agent is a while loop — build one from scratch in the lab, then learn when a workflow beats it and the guardrails production loops need. ⏱ ~12 min lesson · 🧪 ~40 min lab
Lesson 06 · Grounding & proof RAG: Grounding Models in Customer Data Retrieve, then generate: why most RAG failures are retrieval failures, the repair ladder, and a from-scratch pipeline you build and then break. ⏱ ~12 min lesson · 🧪 ~40 min lab
Lesson 07 · Grounding & proof Evals: Proving a Solution Works Success criteria, test sets, and the three graders — code, human, LLM-as-judge — with a lab where you build the eval, then audit the judge itself. ⏱ ~12 min lesson · 🧪 ~40 min lab
Lesson 08 · Grounding & proof Fine-tuning: When It's Worth It Behavior, not knowledge: the decision ladder (prompt → few-shot → RAG → tune), the three honest wins, and a four-scenario decision exercise. ⏱ ~13 min lesson · 🧪 ~25 min exercise
Lesson 09 · Capstone Capstone: Customer Problem → Demoable Solution The full FDE loop — scope, architect, build, prove, demo — run on a realistic freight-company brief, ending in an eval score you can defend. ⏱ ~12 min lesson · 🧪 ~3–5 hr capstone

Reference

How the feedback loops work. Every lesson ends in something you did — a diagnosis, a real API call, scoping math — and includes a ready-made review prompt: copy it into any AI assistant (Claude, ChatGPT, Gemini…), paste your work after it, and you get a senior-engineer review against that lesson's success criteria. Questions mid-lesson? Paste the relevant section into the same assistant and ask. Your progress (lessons, quizzes, labs) is saved in this browser and exportable from the progress card above.