Keith Erik Wilson
In my experience coaching transitioning AI PMs, a strong portfolio is the cheapest way to convert “no AI shipping experience” into job interviews. The format I see working in 2026 is project-led, not resume-led. I find recruiters at AI-first companies increasingly read portfolios before resumes, especially when the resume signals transitioning candidates or first-time AI PMs.
AI Product Manager Portfolio: 7 Projects to Build in 30 Days
In this guide I walk through seven projects I recommend building in 30 days, the portfolio structure I have seen recruiters respond to, how to host and present the work, and how to make the portfolio do double duty as interview content. Build all seven and in my experience you will have months of substantive interview material ready before you apply.
Why a Portfolio Beats a Resume
In a saturated market, recruiters need fast signal of competence. A resume claims experience. A portfolio proves it. Three benefits:
- Portfolio links survive recruiter skim because they show, not tell.
- Strong projects get shared inside hiring teams, multiplying your reach.
- They double as interview content, reducing prep time later.
For candidates transitioning into AI PM from generalist PM, marketing, engineering, or non-tech roles, a portfolio is essentially mandatory. Resume claims of “AI fluency” without portfolio evidence get filtered out within seconds.
Even for experienced AI PMs, portfolios increasingly differentiate. The bar for AI PM has risen sharply since 2023. A portfolio with 7 substantive projects competes effectively even against candidates with 2-3 years of in-role AI shipping experience.
The Portfolio Structure That Works
Host on a simple personal site (Notion, Linktree, Personal domain). Each project gets a one-page write-up:
- The problem you tackled.
- Your approach.
- The eval results or outcome.
- The trade-offs you considered.
- A short reflection on what you learned.
Keep it scannable. Recruiters spend 60-90 seconds per portfolio.
The personal site itself should answer three questions in 10 seconds: who are you, what AI work have you done, and how do I contact you. Everything else is supporting evidence.
A useful template structure: - Hero section: name, headline, photo, one-line pitch. - Featured work: 3-4 strongest projects with previews. - All projects: full list with one-line descriptions. - About: background, current focus, how to reach you.
Project 1: A Custom GPT for a Specific Workflow
Build a Custom GPT in the GPT Store solving a specific PM problem. Examples that work:
- “PRD review buddy” - reviews PRDs for completeness, ambiguity, missing eval criteria.
- “VoC clusterer” - takes raw customer interview notes and produces themes and quotes.
- “AI A/B test designer” - takes a hypothesis and proposes a test design.
- “Stakeholder email drafter” - writes status updates for different audience types.
Document the prompts, the system instructions, and the user feedback you collected. Iterate at least three versions based on feedback.
This project takes 3-5 days. Demonstrates: prompt engineering, product thinking, iteration based on feedback. Shows you can identify a real workflow pain and design an AI solution end-to-end.
Track usage if possible. “200 unique users in week 1” is a strong portfolio signal.
Project 2: An Eval Set for an Existing AI Feature
Pick a public AI feature (ChatGPT, Notion AI, Claude). Design and run a 50-case eval against it. Document what passes, what fails, and the categories of failure.
Structure your eval set: - 30 happy-path cases (typical user inputs). - 10 edge cases (unusual but legitimate inputs). - 10 adversarial cases (prompt injection, manipulation attempts).
Score each case on multiple dimensions: factuality, helpfulness, format adherence, refusal appropriateness.
3-4 days. Demonstrates: eval design, technical fluency, structured thinking. Eval design is the single most under-prepared skill in AI PM interviews. Showing this on a portfolio creates immediate credibility.
Project 3: An AI Product Strategy Memo
Write a 1,500-word memo on the AI product strategy of a public company you understand well. Identify their defensibility, risks, and the next bet they should make.
Structure: - Current state: what they’ve shipped, current positioning. - Defensibility analysis: what’s their moat? - Threat analysis: what could disrupt them? - Recommendation: what should they build next, and why?
2-3 days. Demonstrates: strategic thinking, market awareness, written communication.
This project doubles as interview content. Many AI PM strategy interviews ask exactly this kind of question.
Project 4: A Side-by-Side Model Comparison
Pick a real PM task. Run it across Claude, ChatGPT, Gemini. Document the differences. Recommend which model fits which use case.
Tasks that work well: - “Summarize this 30-minute customer interview transcript.” - “Generate 10 product hypotheses given this customer pain.” - “Write a PRD for this feature description.” - “Critique this go-to-market plan.”
Build a comparison table. Note differences in length, accuracy, structure, refusal behavior, instruction-following.
2-3 days. Demonstrates: technical fluency, structured comparison, decision-making. AI PMs need to make model selection decisions constantly; showing this judgment in a portfolio is high-leverage.
Project 5: A Discovery Synthesis Case Study
Run 5-8 user interviews on a problem you care about. Use AI to synthesise. Publish the resulting opportunity solution tree and the decisions it would inform.
Workflow: - Recruit interview subjects (existing network, Reddit, Slack communities). - Run 30-minute interviews, record (with consent), transcribe. - Use AI tools (Claude, ChatGPT, Dovetail AI) to cluster themes. - Build an opportunity solution tree from the themes. - Write up the resulting product hypotheses.
5-7 days. Demonstrates: research skills, AI synthesis fluency, product judgement. This project is the most valuable interview prep because most AI PM behavior rounds ask about discovery experience.
Project 6: A Pricing Model for an AI Feature
Build a spreadsheet pricing model for a hypothetical AI feature. Include cost per token, expected usage, conversion at three price points, and revenue projections.
Components: - Inference cost calculator (tokens, model, monthly volume). - Tier structure (free, pro, enterprise). - Conversion sensitivity (what if 5%, 10%, 20% upgrade?). - Margin analysis at each tier. - Sensitivity to model price changes.
2-3 days. Demonstrates: pricing fluency, business thinking, modelling skill. Most AI PMs cannot do this confidently. Showing it differentiates immediately.
Project 7: A Public Build-in-Public Project
Build a small AI tool over 4 weeks. Tweet weekly progress. Show the audience the prompts, the eval results, the iteration. End with reflections.
Examples that work: - “I’m building an AI tutor for SQL. Week 1: shipped MVP. 50 users.” - “I’m building an eval framework for content moderation. Week 1: defined dimensions.” - “I’m building a Custom GPT for AI PMs. Week 1: 100 users in first 48 hours.”
4 weeks (parallel to the others). Demonstrates: persistence, iteration, public communication. Build-in-public projects often generate inbound interest from recruiters and hiring managers.
Three Optional Bonus Projects
If you want more projects:
Project 8: A teardown video. Record a 10-minute video tearing down an AI product. Critique it on UX, eval, trust, defensibility. Post on LinkedIn or YouTube.
Project 9: A blog series. Write 4 essays on a focused AI PM topic (eval, pricing, evals, ethics). Publish on Substack or Medium.
Project 10: A community or course. Create a small community (Discord, Slack) for a niche AI PM topic. Lead 4 sessions over a month.
Bonus projects compound the network effect of the portfolio.
The 30-Day Build Plan
| Week | Projects |
|---|---|
| Week 1 | Project 1 (Custom GPT) and start Project 7 (build-in-public) |
| Week 2 | Project 2 (eval set) and Project 4 (model comparison) |
| Week 3 | Project 3 (strategy memo) and Project 6 (pricing model) |
| Week 4 | Project 5 (discovery synthesis) and finalise Project 7 |
End of month: 7 portfolio projects, a personal site hosting them, and 4 weeks of build-in-public public credibility.
This pace assumes 10-15 hours per week of focused work. If you can do more, you’ll finish faster. If you can do less, stretch to 6-8 weeks.
How to Host and Present the Portfolio
Three hosting options:
- Notion: free, fast to set up, easy to edit. Slight whiff of “DIY” but accepted in AI PM hiring.
- Personal domain (yourname.com): most professional. Use Astro, Next.js, or even a simple static template. Buy the domain.
- LinkedIn featured + GitHub + posts: works for individual case studies linked from your profile.
For each project, include: - Title and one-line summary. - Live link or demo. - Source / repo where applicable. - 200-400 word write-up. - Screenshots or short demo video.
The visual presentation matters less than the substance. Clear writing > pretty design.
Using the Portfolio in Interviews
Each portfolio project can become interview content:
- Project 1 (Custom GPT): behavioural questions about iteration based on feedback.
- Project 2 (Eval set): technical questions on eval design.
- Project 3 (Strategy memo): strategy questions on AI defensibility.
- Project 4 (Model comparison): technical questions on model selection.
- Project 5 (Discovery): behavioural questions on research and synthesis.
- Project 6 (Pricing): business questions on AI feature monetisation.
- Project 7 (Build-in-public): questions on persistence, communication.
Before each interview, identify which project is most relevant and prepare to discuss it in detail.
Updating and Maintaining the Portfolio
Quarterly review: - Add 1-2 new projects. - Update or retire stale ones. - Refresh the bio and headline. - Check that all links still work.
Stale portfolios hurt - 2-year-old projects on GPT-3 read as outdated in 2026.
Common Portfolio Mistakes
- Vanity projects with no clear pain solved. “I built a chatbot” doesn’t tell a hiring manager anything.
- Long write-ups without skimmable structure. Recruiters won’t read 2,000-word essays.
- No live links or demos. “Built this” is weaker than “Try it here.”
- Inconsistent quality. Two strong projects beat seven mediocre ones.
- No reflection or learning. What did you learn? What would you do differently?
- Dead links. Test every link before publishing.
- Overpromising. Don’t claim production scale you don’t have.
With 25+ years of experience in Agile transformation, enterprise delivery, and coaching, Keith brings extensive real-world expertise to Agile leadership. Holding 16+ global certifications, including PMP, CSM, and SAFe, he is an Advanced SAFe Practice Consultant specializing in AI-enabled Agile practices, portfolio management, and Scrum Master development. Known for his high-energy, practical approach, Keith brings years of transformation experience across leading organizations
Questions people ask
Published