WahResumefounder

• Role: solo founder: research, product design, design system and front end
• Timeline: 2024 to now. First Figma prototype in 2024, in production since 2025
• Platform: web, desktop first, works down to phone width
• Users: 12,000+

WahResume helps job seekers rewrite the resume they already have so it speaks each job’s language, without the AI inventing anything. I designed and built it end to end. Below are the three decisions that shaped it most.

The problem

Companies need people and still reject millions of resumes before a person reads them. Kathryn Dill’s piece in the Wall Street Journal put a name to what I was living through in my own job search. Applicant tracking systems (ATS) filter on the job’s own skills and keywords, so one well-made general resume keeps losing to people who rewrite for every job.

Rewriting for every job is slow, though. The tools I tried did one piece each: Jobscan scored, builders formatted, chat tools rewrote. Nothing joined them up, and the AI rewriters were happy to make things up.

Who it’s for, and what I heard

I designed for people applying to many roles at once. I talked to recruiters and other job seekers, and spent a long stretch applying myself. Three things stood out:

• They don’t need a template. They have a resume. They need the same true story told in each job’s words, quickly.
• A score needs a next step. A match percentage only helps if it shows what to change and where.
• They don’t trust AI that writes for them. They want to see what changed, and to keep the final say.

That last one became the principle behind every AI feature: an editor, not a ghostwriter.

Process: Figma first, then code

I started in Figma in 2024. I mapped the whole loop as a flow, wireframed the key screens, and built a clickable prototype of the tailoring step so I could test the loop before writing any code.

Once the flows held up, I built a component library in Figma and mirrored it in code with shadcn/ui and Tailwind, so a change to the library had an obvious place to go in the product. From there I worked in short loops: sketch the change in Figma, ship it, watch how people used it, adjust.

Decision 1: one base resume, many tailored copies

The first question was what a “resume” even is in this product. I weighed three models:

1. One master resume you edit for each job. Simple, but every tailoring pass overwrites the last, and the version you sent is gone.
2. A master with live-synced versions. Powerful, but a fix to the master can quietly change a resume you already sent, and the merge rules are hard to explain.
3. A base resume you copy for each job. Each copy is yours to change. Nothing flows back.

I chose the third. You import your resume once and it becomes your base, “the starting point for all your future tailored resumes”. Each job gets its own copy. Tailoring a copy can never damage the original, and replacing the base never touches what you already sent: “Your custom variants are not affected.”

The cost is that a fix to your base doesn’t reach older copies. For people sending dozens of applications, knowing exactly what each company saw mattered more. Each copy sits on the dashboard as a card with a live preview and a status (Draft, Applied, Interviewing, Offer), so the resumes double as a light job tracker.

Creating a copy got simpler over time. In August 2026 it asked for a starting layout, a name and an optional job profile before you could start. Two weeks later the name went, and by late September it asked for one thing: “Paste the job description you’re applying for.” If you skip it, it doesn’t block you. It explains why it helps and lets you carry on.

The editor itself is meant to feel like writing, not filling in a form. It works like Notion: type “/” for resume blocks (Experience, Education, Skills), drag blocks by their handle, and select any text to “Ask AI” to improve it, fix the grammar, or make it shorter or longer. The AI’s suggestion appears beside your text with three choices: “Replace selection”, “Insert below” or “Discard”. Your words are never swapped out without a click.

Decision 2: a score you can act on

Most resume scanners give you a percentage and a list of missing keywords. Neither tells you where a keyword should go, or how to say it.

I split the problem in two. The WahResume Score is half resume quality and half job match, and it updates live as you type, so every edit shows its effect. It reads in three plain bands, “Needs work”, “Needs polish” and “Strong”, and 80 is the line recruiters shortlist from.

Below the score, the job description is shown as it was written, not boiled down into a keyword cloud. Every skill and keyword in it is highlighted in place: skills in blue, keywords in purple, solid when your resume has them, dashed when it doesn’t. Each section counts what’s left, as in “X of Y skills, A of B keywords in your resume”, and a requirement gets a check once it’s covered. You can see what the job is asking for and how much of it you’ve already said.

Then I made each term clickable. Click a missing term and you see where it already appears in your resume, if anywhere. Ask it to “Suggest a placement” and the AI picks one line, in your summary or one of your three most recent roles, and rewrites as few words as possible. The change previews right in your resume, crossed-out words and added words, with “Apply” or “Suggest another placement”. A line underneath says “Only add this if it’s true for you.” Applying shows an Undo.

I ruled out two easier options. A checklist alone leaves the hard part, the wording, to the user. Auto-inserting every missing term produces keyword stuffing that recruiters spot straight away. The placement flow does one term and one line at a time, and you approve each one.

Decision 3: Enhance as an editor, not a ghostwriter

Placement fixes one term at a time. Some resumes need a bigger pass, so I designed “Enhance”. The risk is obvious: a one-click rewrite is exactly the ghostwriter people said they didn’t want. So the design is about control at every step.

• You choose what changes. A short picker asks “What should AI improve for this job?”: rewrite the summary, rewrite recent experience, add missing skills and keywords, match your job titles, tighten bullets, and more. Each option shows its likely effect on the score (“up to +N”), and only the fixes your resume needs start ticked.

• You see the work. While it runs, it names its steps in plain words, ending with “Checking every change against your resume”.
• You compare before you commit. The result opens as a draft with an “Original | Enhanced” switch and the old score beside the new one. You can edit the enhanced version before accepting, and nothing is saved until you choose “Accept” or “Discard”.
• You’re told what to check. A note says exactly what the AI added. If it estimated a number from your role, it says so and asks you to swap in your real figure.

Behind the interface are rules the AI can’t break: use only numbers that appear in your resume, never claim a seniority your latest role doesn’t show, never move a detail from one job to another. If a rewrite slips in a number you didn’t confirm, it’s thrown away, and the app says so: “The AI tried to add numbers you didn’t confirm, so we threw those changes away.”

To know whether Enhance was any good, I didn’t rely on the score alone. I wrote a ten-point quality rubric (no invented numbers, at most one added term per bullet, faithful to the source almost every time) and checked outputs against it. A model from a different family judged them blind, and I rated them myself.

One brand, the site and the product

People meet WahResume on the marketing site and are in the product a minute later. It should feel like one product, not two.

So in June 2026 I moved the design system into one shared package that both use. It holds three things:

• Tokens: colours, corner radius, shadows, and light and dark mode.
• Type and motion: Geist for the product, Sora for the site’s headlines, and the small animations people see everywhere, like the scan line and the score filling up.
• Components: 28 building blocks, from buttons and badges to dialogs and tabs.

The Figma library uses the same tokens, so design and code stay in step.

The product takes its components straight from the package. Change the button once and every button in the product updates. Every core component has a Storybook story with accessibility checks. A few product choices sit on top, like warmer greys for the editor and a neutral hover, so green stays a brand colour and not decoration. On top of the shared set, the product has its own components for the jobs only WahResume does.

The site shares the same colours, type, logo and dark mode, and uses shared pieces such as the share button. Moving the rest of it onto the package is next.

Accessibility was built in from the start: keyboard navigation, visible focus, and focus that stays inside open dialogs.

Results

WahResume has 12,000+ users. It now covers the full loop: a base resume, tailored copies with live scores, one-line keyword placement, Enhance with review, cover letters that cover the terms the resume misses, and mock interview practice.

What I’d do next

The product still has gaps, and they’re the next things I’d design:

• Accept changes one at a time in Enhance. Today it’s accept all or discard all. The engine already tracks each change; the review screen should let you keep some and drop others.
• One consistent signal for AI. AI actions currently use more than one icon and colour. A single, consistent marker, as a design token both surfaces share, would make it obvious what the AI touched.
• Version history. Copies protect the base, but within a copy, the only undo is the editor’s own.
• Save status on desktop. It shows on phones but not on larger screens, where most editing happens.