How this works

So you can make it better! This is a simple website overall, and there's a lot of room for improvement. For anyone who needs a project to improve or advise on to improve your portfolio, this is here for you to adopt.

So take it 👍 The site is four files with no build step, and a fork publishes to your own GitHub Pages URL in about two minutes. The Customizing Runway section of the README walks through all of it, including what the nine color variables at the top of the stylesheet control.

Share your customizations and changes! Post it on LinkedIn and send it to me. I'll build a library to show what people have done. So share Runway with people you know and build your portfolio with me.

• You can inspect how it works and see what the AI sees.
• Your portfolio and résumé never reach a server, because there is no server.
• You pick the model and use your existing environment.

You can use ChatGPT, DeepSeek, decide between Claude Sonnet or Fable, whichever you prefer. Search through the prompt, customize it, paste it, what you use is yours.

AI tools are still data-driven systems and are not a true replacement for a human coach or perspective. To best understand any AI output, it's valuable to know the process that built it. By having the prompt in your hands, you can pick apart how it works, learn from it, and experiment with tweaking it.

Runway decides specific titles under what you select. For example, "Design" includes UI/UX, Graphic, Design, and Industrial Design. Sponsorship data is public and Runway makes it presentable. The same data produces the same shortlist every time, and every number traces to a filing.

Your LLM decides how to build the report. That half is judgment, and it is only as good as the model you run it in and the materials you give it. Read it critically. It is a well-briefed opinion, not a guaranteed answer.

I recommend experimenting with different models if you can. With Claude, see if Sonnet or Opus gives you different recommendations. You can even have a sub-agent run an adversarial check against a report to see if multiple agent perspectives yield new insight.

The prompt hands your LLM two things: aggregate patterns across every sponsor in your role - which job titles recur, which occupations dominate, whether they hire in-house - and your own materials.

The patterns decide which skills matter. Your materials decide which of those you already show. That is why each skill in the report carries both a market reason and a line pointing at something in your own work: a rating you cannot trace is one you should not trust.

If you want to further fine tune your recommendations, I recommend reading through the prompt in something like Notion or Obsidian and tailoring it to fit yourself better. You can experiment with customizing the prompt before you use it.

There are so many possible things to work on. One of the hardest parts of job searching is figuring out what will actually help. Job hunting generates more advice than anyone can act on.

The report commits to one direction and says why. If you disagree with the reasoning, that is useful too. Hopefully it helps you find something to start on.

From personal experience, working on too many things can slow down how well you can showcase your strengths. Deeply investing in a project reflects how valuable work often goes in a professional environment, and it helps keep you from spreading yourself too thin.

A way to standout based on patterns in capabilities of recognized people in your role tend to share. Qualities that differentiate real standouts from equally qualified peers. It is deliberately harder than the main recommendation, and it is about something to do to even further standout.

It is identified when at least three proven people in the field visibly share it. The recurring pattern is the evidence, not any one person's story. It has to be a craft you can build toward, never a job title you land or a company you found. The standout people are the receipts so you can trace a recommendation and learn more about them and what made them identifiable as remarkable people. The report recommends the highest-ranked path you do not already have and shows the other identified paths so you can see the whole picture.

Because AI should show where it sources its info. Every report ends with a record of whether Runway actually read your portfolio and résumé, and whether it researched the companies.

A report built without your portfolio is inference, not observation. Saying so is the difference between advice you can weigh and advice you have to take on faith.

The company list comes from US Department of Labor LCA disclosure data: the filings an employer must make public to sponsor a work visa. Runway keeps only certified filings at the entry wage (Prevailing Wage Level I) whose occupation codes match the role you picked, then groups them by employer.

The funnel line under the table shows every number that survived each step, and the provenance link next to it gives you the raw source. Nothing in the shortlist is estimated or inferred.

No. A certified LCA means an employer sponsored this occupation at this wage level at some point - not that a role is open today, and not that they will sponsor you. Treat the shortlist as "has sponsored before," which is still far better evidence than a list of companies that merely sound international.

The caveats are in the report as a reminder.

Nothing leaves this page: no upload, no account, no database, no server to send anything to. The site is static files.

Your résumé is a file path, not the file: Runway never opens it. The path travels into the prompt text so that an LLM with file access can read it when you run it.

The advice layer is a language model reading your work: it can misjudge what a case study proves, especially the difference between a polished concept and something real people used. Advice is meant to be a logical perspective based on data the model finds. Logical people can be wrong too, and right/wrong is what you make of it.

Several things weaken accuracy. Some roles have thin data: a narrow role in one quarter may rest on a few dozen filings, and small samples mislead. Runway normalises employer names conservatively, so one company can appear twice.