Five scores beat one

Upload a resume. Get “78.” Now what?

Did ATS parsing fail? Are skills buried? Does the tone miss the role? One number feels decisive and teaches nothing, so people rewrite blindly and apply again.

I built Resumai to refuse that shrug.

I worked solo, end-to-end, and started where trust is won or lost: the analysis view.

Five dimensions (ATS, tone, content, structure, skills), each 0-100, each with three to four concrete tips. Optional job description for role-aware feedback. History so you can iterate versions.

Only after that screen felt useful did I wire PDF upload (client-side PDF.js), Puter auth/storage, and the rest of the shell. Login exists to serve the loop, not the other way around.

A hosted LLM shipped a useful product fast. The cost is coupling quality and spend to a provider, acceptable for an MVP, something to revisit if volume grows.

Client-side PDF parsing keeps files off my server and fails on exotic PDFs. Five scores take more UI space than one badge. That space is the product: if you cannot explain the grade, you have not helped anyone.

Try the public demo: analyze → tips → history. Stack: React Router, TypeScript, Tailwind, Puter.js, Zustand, PDF.js.

If you hire for people who can own a thin AI product surface, or you need a similar MVP, start with the live demo, then the Resumai case study, then the contact form on this site.