Notes

Artificial Intelligence

I Made It Easier for Recruiters and Hiring Managers to Review My Portfolio

A small feature to make portfolio reviews easier for hiring managers and recruiters, and what I learned testing it against real job descriptions.

Michael Whitaker6 min read

Dark-mode homepage actions: a white “See my work” button with an arrow, and beside it a quieter “Short on time? Copy for your agent” action with a robot icon shaped like a copy symbol.

Figure 1. The final homepage design in dark mode: “See my work” as the primary action, with a quieter secondary action that copies the prompt for an AI agent.

I had a simple thought while working on my portfolio.

A lot of hiring managers and recruiters are already using AI tools to summarize resumes, compare candidates, and make sense of job descriptions. So instead of pretending that is not happening, I thought it might be useful to make that process easier and a little more honest.

So I added a small feature to my site: a Share with your agent button.

It copies a prompt that tells ChatGPT, Claude, or another assistant to review my public résumé and portfolio, then compare them against a job description. The assistant is supposed to say what it could actually retrieve, ask for the job description, and only then assess fit.

The problem I was trying to solve

This little feature came from a real friction point.

If someone lands on my site, they still have to do a fair amount of work:

  • Scan my résumé
  • Read case studies
  • Figure out what is relevant to their open role
  • Decide whether the evidence is strong enough to warrant a conversation

That is normal, but if they are already using AI to help with hiring workflows, I can at least make that step easier.

I also liked the idea that the feature itself could express something about how I work: product thinking, AI literacy, design-to-code workflows, accessibility, content structure, and iteration based on live testing.

What I built

The first version was a small banner below the main navigation, with a robot icon, a short sentence, and a button. I put it there so the feature would be visible no matter which page someone landed on, which mattered for busy recruiters and hiring managers.

Light-mode portfolio navigation with a teal banner and a Share with your agent button.

Figure 2. The initial global-banner design, just below the site's primary navigation.

Clicking the button copies a structured prompt to the clipboard.

That prompt tells the assistant to:

  1. Retrieve my public résumé and portfolio.
  2. Report what it actually retrieved.
  3. Ask for the job description.
  4. Compare my published experience against the role.
  5. Distinguish documented evidence from inference or uncertainty.

I also created public, agent-readable versions of key materials:

  • resume.md
  • portfolio.md
  • llms.txt

The goal was not to dump my whole website into a prompt. The goal was to give the assistant a clean, reliable path to the evidence.

The llms.txt index listing the agent-readable résumé and portfolio, the published case studies, the website pages, and notes for agents.

Figure 3. The llms.txt index that points assistants to the public Markdown résumé and portfolio.

What didn't work the first time

The first version was more complex and much longer. GPT over engineering as models like to do. Technically, everything was fine. The files were public. The site served them correctly. The button copied the prompt. The UI worked, but during live testing, in some ChatGPT runs, the assistant could read some of the markdown, but failed to retrieve the résumé and portfolio. In GPT, Claude, and Gemini runs, the assistants also tried to proceed with too little evidence. So I treated it like a product problem, not just a technical one.

What I changed

I ended up simplifying both the delivery and the instructions.

A few of the more important changes:

  • Served the public portfolio files as plain text for better compatibility.
  • Shortened the copied prompt substantially.
  • Made the prompt focus on retrieval first.
  • Added an evidence gate so the assistant should not make a recommendation without enough source material.
  • Made it explicitly separate documented evidence from assumptions.
  • Tested it repeatedly in ChatGPT and Claude using real job descriptions.

That shorter prompt worked much better.

In two rounds of testing with different job descriptions, both ChatGPT and Claude retrieved the complete documents and produced comparable assessments, including similar observations about strengths and missing evidence.

What the testing was actually useful for

There were two outcomes, and both were valuable.

The first was the obvious one: it helped test whether a hiring manager or recruiter could get a useful AI-assisted read on my fit for a role.

The second one was probably more valuable, but also something I already suspected and am working on.

By running real job descriptions through the feature, I got a clearer view of where my public evidence is strong and where it is thin.

AI assistant assessment showing retrieved portfolio evidence and a role-fit evaluation against a job description.

Figure 4. A live Claude evaluation using the public résumé, portfolio, and a real job description.

For example, the tests consistently picked up strengths around:

  • Enterprise product design
  • Design systems
  • UX Engineering collaboration
  • Modernization work
  • Complex workflows
  • Cross-functional leadership

They also consistently surfaced areas where my public portfolio is lighter than my actual experience, especially when it comes to:

  • Measurable outcomes
  • Some hands-on product examples
  • Alerting and notification patterns
  • Mobile-specific evidence
  • Certain workflow and research details

That was useful. It helped separate actual experience gaps from documentation gaps.

A small feature with a double benefit

So the feature now does two jobs.

It helps make my portfolio easier to review against a role.

And it helps me audit the portfolio itself.

That second part was not the original point, but it turned out to be one of the more useful outcomes.

Why it belongs on my portfolio

I've been using AI-assisted workflows to plan, design, develop, and validate this portfolio. Adding this feature was another opportunity to put some of those ideas to work.

It involved thinking through the interaction, structuring the source material, accounting for accessibility and failure states, and testing actual retrieval behavior across multiple assistants.

It is an extension of the product design and development work I've already been doing, with my own website serving as the product.

One more design iteration

After the feature was built, tested, and live, I kept looking at it.

The banner worked exactly as intended. It made the feature immediately discoverable, regardless of which page someone landed on.

But the more I looked at the homepage, the more I questioned whether that was really the right place for it.

The homepage has a fairly deliberate introduction, with oversized typography, motion, and a little drama. The banner wasn't necessarily bad, but it interrupted that experience and gave the AI feature a level of prominence I wasn't sure it needed.

I explored moving it into a two-column section below the introduction, alongside the existing "See my work" action. That felt closer, but subsequently seemed like more UI than the interaction deserved.

The final solution was much simpler. Keep "See my work" as the primary action and place a quieter secondary action beside it: "Short on time? Copy for your agent."

I also worked a little visual shorthand into the icon. It's a robot, but its shape doubles as a copy symbol. A subtle background fill on hover makes the entire control feel interactive without introducing another prominent button.

The original banner still exists in the codebase, but it's no longer rendered. The functionality is the same. The way it fits into the experience is different.

What started as a small feature to help recruiters and hiring managers also became a reminder that getting something working and finding the right place for it aren't necessarily the same thing.

What's next

The feature is live, but it also pointed to the next body of work.

I need to strengthen the public evidence behind it by adding or expanding a few things:

  • More substantive product case study material
  • Notes on selected projects or workflows
  • Clearer outcome language where I can support it
  • Better public coverage of work that currently lives mostly in résumé bullets or protected case studies

So in that sense, it made the experience better and reinforced what to improve next.