AI Engineering from Scratch: choose a route through the curriculum
How the curriculum connects math, models and agents
Read the phase graph as a dependency map, then test where your prior knowledge lets you enter.
What you will learn
- Follow the prerequisite chain
- Understand a lesson’s internal loop
- Make the route measurable
Before you start
- Basic command-line access
- A bounded learning goal and disposable checkout
Turn lesson commands and code changes into a reviewable record of what you can do.
Key takeaways
- The phase graph has branches and prerequisites.
- Lesson files are the inspectable unit of learning.
- A useful progress record ties an artifact to a concept.
Follow the prerequisite chain
The README draws a path from setup through mathematical foundations, model construction and LLM engineering to tools, agents and production. It branches for vision, speech, reinforcement learning and other topics instead of presenting every phase as one linear track.
This map is an editorial guide to prerequisites. It does not enforce progression in software. A learner can enter at agent engineering, then return to attention or optimization when a concrete problem exposes a gap.
Understand a lesson’s internal loop
A typical lesson has explanatory text in docs/en.md, runnable code, an output artifact and a quiz. The repository tells readers to run from the root and preserve command evidence. Those files are the unit to inspect when a course claim seems vague.
The website and book provide alternative reading surfaces over the same curriculum. The language-specific landing pages are translations, while the i18n document warns that some lesson translations live on a separate branch; check the file you actually opened.
Make the route measurable
Pick one concept you want to explain and one artifact you want to create, such as an agent loop trace. Link the artifact to the phase and lesson path that taught it. A long streak of completed pages is a weak substitute for that link.
At each branch, list the prerequisite you used and the unresolved idea. That prevents a later capstone from becoming a bundle of copied commands with no account of why they work.
Decision guide
| Criterion | Option A | Option B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
Implementation steps
- 1
Draw your selected phase path and its prerequisites.
- 2
Open one lesson’s docs, code and quiz.
- 3
Attach a runnable artifact to the concept it demonstrates.
Copy-ready example
goal: explain one agent-loop failure
prerequisite: LLM application basics
lesson: phases/14-agent-engineering/01-the-agent-loop
evidence: command, output, one explained code changeFrequently asked questions
Is the website a separate course?
The README points to the same curriculum through GitHub and the website.
Are all translations equally current?
Check docs/i18n.md and the exact lesson file; the README distinguishes landing pages from lesson translations.
Sources
- AI Engineering from Scratch / README.mdSource checked 2026-09-29
- AI Engineering from Scratch / book/README.mdSource checked 2026-09-29
- AI Engineering from Scratch / docs/i18n.mdSource checked 2026-09-29