A portfolio that shows judgement

An AI engineering portfolio is the work you show an employer, and it should show how you decide. Two people each show a chatbot. One shares a screenshot. The other shares an eval table, a list of known failures and a week's running cost.

A hiring manager learns far more from the second. This lesson is about building that second kind of portfolio from your capstones.

A visitor gives your repository a minute or two. The README has to carry that minute.

Say what the project does in one sentence. Say why you built it, and who it is for. Then say how to run it in a few steps, with fixed sample data, so it is reproducible. Put the eval results near the top, not at the bottom.

Anyone can claim a feature works. Evidence is a number someone can check: an eval score, a cost, a latency.

Known limits matter just as much. Saying where your project fails shows that you tested it. It also shows you know what to fix next. The program checks a README's headings against a checklist.

# The headings found in a capstone README.
headings = ["What it does", "How to run it", "Eval results"]

checklist = [
    "What it does",
    "Why I built it",
    "How to run it",
    "Eval results",
    "Known limits",
    "What I would do next",
]

missing = 0
for item in checklist:
    if item in headings:
        print("ok      ", item)
    else:
        print("missing ", item)
        missing = missing + 1

print(missing, "of", len(checklist), "missing")
# Add "Known limits" to headings, then press Run again.

Before a repository goes public, check three things. No secrets: no keys or passwords, anywhere in its history. No personal data: use made-up names and sample documents.

And it is yours: say which parts an assistant wrote, and credit any code you reused. Add a licence, so others know what they may do with it.