Learn to build with AI

One request, drawn end to end, in five parts. Each part is one of the courses.

Fig. 1 — one request, end to end

The request travels hereOnly if the model calls a toolWatching — carries nothing

5courses86lessons6–8min a lesson613min in total

Figure 1: how one request moves through the machine

A prompt goes in — a system prompt, then the user’s turn.
It is cut into tokens.
Everything meets at the context window. Retrieved passages arrive there too: the prompt goes back out as a query to a corpus, candidates come back, they are ranked, and only the top few are kept.
The model runs — a stack of layers, with attention running across the tokens inside each layer.
The model either stops, or calls a tool. A tool result comes back round into the context window and the loop runs again.
An answer comes out, in the shape the prompt asked for.
The output is tapped on the way past, traced and scored — and so is what retrieval brought back.
Two things the drawing does not show: retrieved passages and tool results are counted in tokens too, and the context window’s limit covers what comes out as well as what goes in.

The five parts of Figure 1, as links to the courses that teach them

Key to Figure 1 — the five parts, and the course that teaches each

Prompt
What you send in, and the shape you demand back.
Prompt Engineering
Instructions that hold up in production.
19 lessons · 133 min
Retrieval
What gets fetched and stuffed in beside it. Only the top few survive ranking.
RAG
Retrieval that actually retrieves.
23 lessons · 169 min
Model
The thing in the middle, and which one to choose.
LLMs
The model layer, and how to choose in it.
18 lessons · 126 min
Tool loop
The loop out to tools and back.
Agents and Tool Use
Giving a model hands, and keeping them on a leash.
14 lessons · 103 min
Measurement
Whether the output was any good, and whether retrieval found the right thing. A trace is one run, recorded.
Evals and Observability
How to know it is working, and how to know when it stops.
12 lessons · 82 min

Notes

The bits nobody writes down, in the order you actually need them. Free, hands-on courses for engineers who can already code but have not built with LLMs. Lessons are short and stick to the real mechanics — what the model is doing, why it does the odd thing it just did, and what to reach for instead. Each one ends with a quick recall quiz.

What the drawing leaves out: retrieved passages and tool results are counted in tokens too, and the context window’s limit covers what comes out as well as what goes in.

Everything here is free and stays free. Your progress is kept in this browser only — no account, nothing sent anywhere.