AI Engineering from Scratch: choose a route through the curriculum
Build a learning portfolio from the course’s artifacts
Turn lesson commands and code changes into a reviewable record of what you can do.
What you will learn
- Select a capstone with a narrow claim
- Package the evidence
- Close with unresolved limits
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
- A portfolio entry needs a bounded claim.
- Source and original work should be distinguishable.
- Unresolved failures belong in the record.
Select a capstone with a narrow claim
The curriculum ends with capstone projects, but a useful portfolio entry can begin earlier. Choose one behavior, such as an agent’s tool-call retry, and state the input, expected outcome and failure boundary.
Keep the original lesson path and license notice beside your own changes. Make clear which code comes from the course and which part you designed or modified.
Package the evidence
Include the pinned revision, setup commands, dependency versions, test input, output and one failure you investigated. If a provider was used, record the model and configuration without exposing credentials.
The course’s book and website help readers navigate, but a portfolio needs runnable files or a trace another person can inspect. A screenshot alone cannot reproduce an agent decision.
Close with unresolved limits
Ask a peer to rerun the smallest example and point out where instructions fail. Fix that path before claiming the exercise works on other platforms. If the result depends on paid access, say so.
This article proposes a portfolio structure; it does not claim the repository generates one automatically. The finished artifact should show a specific skill and the conditions under which it was observed.
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
Choose one small behavior to demonstrate.
- 2
Save source attribution, setup and a failed case.
- 3
Ask another reader to reproduce the result.
Copy-ready example
portfolio_entry:
source_lesson: phases/14-agent-engineering/01-the-agent-loop
revision: bf7791e140768d8223d24e616bb60cbf07fea014
own_change: describe
reproducible_run: pending
known_failure: recordFrequently asked questions
Does the repository publish my portfolio?
No. This is a reader exercise assembled from lesson artifacts.
What is a convincing first entry?
A small runnable example with clear inputs, output and an explained change.
Sources
- AI Engineering from Scratch / README.mdSource checked 2026-09-29
- AI Engineering from Scratch / book/README.mdSource checked 2026-09-29