AI Engineering from Scratch: choose a route through the curriculum
Read the preflight source before trusting a green check
Trace a small Python utility and learn what its result can and cannot establish.
What you will learn
- Start at the command-line entry
- Follow one failure path
- Keep observations separate from inspection
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 checks a selected route.
- A green preflight does not execute all lessons.
- Source inspection and runtime observation are different evidence.
Start at the command-line entry
The README invokes verify.py with a route argument. In the fixed source, trace argument parsing, each environment probe and the final exit status. Write down which checks are required for the selected route and which are advisory.
A check can report that Python or a command-line tool exists while a later lesson still fails on package versions or permissions. Treat the preflight as an early diagnostic, not a certificate that every exercise will run.
Follow one failure path
Choose a harmless missing-tool scenario in a disposable environment. Compare the printed reason with the branch that produced it and the corrective instruction. This is a source-reading exercise, not an invitation to remove system software from your daily machine.
The Docker lesson’s Dockerfile is another bounded reading target. Identify its base image, copied files and execution command, then relate those choices to the exercise rather than projecting them onto the entire course.
Keep observations separate from inspection
This series inspected source text at a fixed revision but did not execute verify.py or the Docker lesson. A reader’s runtime receipt should record its own platform and result beside the source path.
The useful output of the trace is a table of checks, required routes and observed outcomes. Do not infer reliability from a successful exit on one workstation.
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
Locate route parsing in verify.py.
- 2
Trace one success and one failure branch.
- 3
Compare the source contract with your own recorded run.
Copy-ready example
input: --route beginner
trace: argument parsing -> environment probes -> severity -> exit code
record: source branch + actual output on your machineFrequently asked questions
Does verify.py install missing tools?
Read its source and output for the selected route; do not assume the diagnostic also changes the machine.
Did this review test its exit codes?
No. The file was inspected at the pinned revision.
Sources
- AI Engineering from Scratch / README.mdSource checked 2026-09-29
- AI Engineering from Scratch / phases/00-setup-and-tooling/01-dev-environment/code/verify.pySource checked 2026-09-29
- AI Engineering from Scratch / phases/00-setup-and-tooling/07-docker-for-ai/code/DockerfileSource checked 2026-09-29