An AI skills portfolio should hold evidence an employer can check, not claims. Build it in six sections: a short summary, a sample of review decisions with reasons, one or two improvements you proposed, certificates with verification links, references, and what you are looking for. Keep it to a few pages or one web page, and anonymise everything.
What is a portfolio for in AI operations?
A CV says what you did; a portfolio shows how you think. For people moving into AI operations, that matters because the job is judgement: approving, amending, rejecting and escalating what an agent proposes. A hiring manager wants to see your reasons, not just your job titles. If you are new to the idea, the guide to what a skills portfolio is covers the basics.
Portfolios also help people whose experience does not fit a standard CV: career changers, contractors, people returning after a break, veterans. The guides on how contractors and military veterans can build a portfolio employers will trust go into those situations.
Where does the evidence come from?
Four sources, in order of strength:
- Real review decisions from a pilot or your job, anonymised and with permission.
- Practicum or lab decisions, marked against a standard.
- Practice decisions from a simulated queue, logged with reasons.
- Written analysis, such as a proposal to change an agent's limits based on cases.
Use the strongest you have. Never include customer data, client names or anything your employer treats as confidential. If in doubt, ask your manager before including a work example.
The portfolio template
| Section | What to include | Evidence an employer can check |
|---|---|---|
| 1. Summary | Three lines: who you are, the decisions you review or run, the role you want | — |
| 2. Decision samples | Five to ten anonymised decisions: what the agent proposed, your verdict, your reason | Practicum sign-off or manager reference confirming the work |
| 3. Patterns and improvements | One or two cases where you spotted a pattern and proposed a fix to a brief, source or limit | Note from the agent's owner or pilot lead |
| 4. Certificates | Each certificate with issuer, dates and status | Verification link or number |
| 5. Learning record | Courses and labs completed, with dates | Badges or completion records |
| 6. References | Two people who saw your review work | Contact details on request |
| 7. What next | The roles you are applying for and where you can work | — |
How do you write a decision sample?
Use the same four lines for each, so a reader can compare them quickly:
- Case: one line describing the type of case, without identifying details.
- Agent proposed: what the agent drafted or wanted to do.
- My decision: approve, amend, reject or escalate.
- Reason: one or two sentences, including the source or rule you relied on.
Choose a mix. Include at least one approval (to show you do not reject by reflex), one rejection and one escalation. Reviewers who only show catches look as if they are hunting for errors rather than judging cases.
A worked example
A section 2 entry for an invented operations analyst, Wei, from a practicum on a simulated payments queue:
Case: request to pay a supplier invoice; supplier's bank details changed in the past week. Agent proposed: hold for review, flagged as a new payee. My decision: reject until verified. Reason: the bank change arrived only by email, which is an external source. I asked for confirmation through the contact on file before any payment.
Case: customer refund request for a damaged item under the refund limit. Agent proposed: approve refund. My decision: approve. Reason: order record and delivery photo both support the claim; within policy and limit.
A section 3 entry from the same portfolio:
Pattern: six held refunds in a week were for the same product line, all within limit but held because the returns policy source was out of date. Proposal: update the returns policy source; no change to limits needed. Outcome: accepted by the queue owner in the simulated workspace.
How should you present it?
One page on a personal site, a shared document or a PDF with live links all work. What matters is that links go to sources an employer can check: a verification page for each certificate, not an image. A portfolio career built on freelance or contract work can use the same structure across clients, with each client's permission.
Update it after every pilot, practicum or new role. A portfolio with dated, current certificates is stronger than a large one with lapsed ones. For a step-by-step build, see how to build an AI skills portfolio.
Questions people ask
How long should the portfolio be?
Short enough to read in ten minutes. Five to ten decision samples and one or two improvements are plenty; more dilutes the best ones.
Can I include decisions from a simulated queue?
Yes, labelled as such. Simulated decisions marked against a standard are real evidence of judgement, and an employer will respect the honesty of the label.
Do employers actually read portfolios?
Not all of them. Link it from your CV and LinkedIn, and bring two samples to interviews. The AI skills for job seekers guide explains what employers ask you to prove.