AI Engineering from Scratch: choose a route through the curriculum
First lab: run the preflight and keep a useful record
Get a reproducible first result without installing the entire AI tooling stack.
What you will learn
- Keep the first setup small
- Read the failure before installing anything
- Save a lab receipt
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 preflight has route-specific requirements.
- A small dependency-free exercise can establish the workflow.
- Secrets do not belong in a shared lab receipt.
Keep the first setup small
The README begins with a repository clone and a Python preflight. The script separates tools required for the selected route from tools needed later, so a missing GPU or container engine does not automatically block a beginner lesson.
Run commands from the checkout root as the lesson instructions require. Record Python version, operating system and the exact revision. Those details explain many differences between a reader’s output and a screenshot from the course website.
Read the failure before installing anything
The inspected verify.py is source for the preflight, not evidence that it passed on your machine. If it reports a missing dependency, use the corrective command for the route you chose and rerun only that check.
The linear algebra example cited by the README is dependency-free. Its small matrix-vector calculation is a better first success criterion than downloading model weights or configuring several provider keys at once.
Save a lab receipt
A useful receipt contains the working directory, command, exit code, meaningful output and one modification you can explain. A copied terminal transcript with no interpretation shows execution, but little understanding.
Do not paste API keys into a public learning log. If you later use the tutor installation path, first review which agent host and project or user scope the installer will modify.
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
Clone the repository and pin the revision in your notes.
- 2
Run verify.py with the route you chose.
- 3
Run one lesson and record its output and your small change.
Copy-ready example
receipt:
revision: bf7791e140768d8223d24e616bb60cbf07fea014
route: beginner
command: python3 phases/00-setup-and-tooling/01-dev-environment/code/verify.py --route beginner
exit_code: record-after-runningFrequently asked questions
Does a failed optional tool check stop all lessons?
Check whether the selected route requires that tool; the preflight distinguishes immediate from later needs.
Was this preflight executed for the article?
No. The commands come from the fixed README and source and need local verification.
Sources
- AI Engineering from Scratch / README.mdSource checked 2026-09-29
- AI Engineering from Scratch / phases/00-setup-and-tooling/01-dev-environment/docs/en.mdSource checked 2026-09-29
- AI Engineering from Scratch / phases/00-setup-and-tooling/01-dev-environment/code/verify.pySource checked 2026-09-29