Resumind
An AI resume analyzer that scores a CV against a specific job description and returns structured, actionable feedback — running auth, file storage, and model inference entirely client-side.
Fig. 01 — Resumind interface
01 — Problem
Generic resume checkers grade you against nothing in particular. You get a number and no idea what to change. I wanted the opposite: paste the job you actually want, get feedback tied to that posting. The catch was cost — an AI tool with per-user file storage and inference normally means a server, a database, and a bill that grows with every visitor.
02 — Build
No backend, on purpose
Puter.js handles auth, file storage, and model inference from the browser, billed to the end user rather than to me. That removed the entire server tier: no API routes to secure, no keys to leak, no database to run. The whole app deploys as static output.
A scoring pipeline, not a single prompt
Feedback comes from structured prompts that return typed JSON — ATS compatibility, tone, content, structure, and skills, each with its own score and specific fixes. Parsing into a schema instead of rendering raw model prose is what makes the results renderable and comparable across runs.
A dashboard that survives real use
Uploads accumulate fast, so the tracker got search, sort, archive and restore, inline editing, and one-click export of feedback to a text file. Global state runs through Zustand — a small store beat prop-drilling across the upload, results, and dashboard routes.
Fig. 02 — Resumind — additional view
03 — Decisions & tradeoffs
- Client-side inference over a Node API layer
- Kept hosting free and stateless. The tradeoff is a hard dependency on an external SDK and no server-side rate limiting — acceptable for a tool where each user brings their own quota.
- Typed JSON responses over free-form model output
- Made the UI deterministic. A malformed response fails one card instead of blanking the page.
- React Router v7 over Next.js
- The app is genuinely client-heavy — there is no SSR to gain when every operation happens in the browser. Less machinery for the same result.