Matchify AI
One resume, one job description, a score the model does not invent.
Overview
Upload a PDF, paste a job description, and get a score, matched skills, missing skills, and a few rewrite ideas. The number is calculated in code: 50% must-have skills, 20% nice-to-have, 20% keywords, 10% experience. Reports are saved to the account.
The Problem
A model that returns the score will invent one. A job seeker still needs to see which skills matched and which did not.
Key Features
- PDF plus a pasted job description, with progress over server-sent events
- Score computed in lib/scoring.ts, not by the model
- Keyword tools and bullet rewrites, then a second score
- Google, GitHub, and email login. Saved reports, 10 per page
- Resume and job text treated as untrusted input
- Same resume and job description within an hour reuse the cached draft
Technology Stack
My Contribution
Designed the agent loop, kept scoring in application code, and stored analyses per account.
Challenges
No OCR. A scanned PDF comes back as unreadable. PDF only, at most 5 MB and 10 pages.
Outcome
A public sample of an analyzer where the weights are fixed and tested, including empty, full match, miss, and a few synonyms.
Interested in working together?
Tell me about your project and let's see if I'm a good fit.