Upwork Proposal Generator for AI Engineers

Chasing a 90 percent ATS score is where most of these builds quietly break. The scanners you're targeting (Workday, Greenhouse, Taleo, iCIMS) each parse resumes differently, so a resume that scores 90 in one parser can drop to 60 in another. The real engineering problem isn't getting Claude to swap in keywords, it's mapping the job description to the right skill vocabulary without stuffing terms that get a candidate flagged as gaming the system, and keeping formatting the parser can actually read.

I build AI systems on exactly this stack. Two that are close to your problem:

PerfectMatchSchools (perfectmatchschools.com): a live ATS-style platform I built solo where teachers get matched to schools through a scoring algorithm. Before, matching was manual and vague; after, applicants got ranked fits automatically. Same core discipline your resume-scoring engine needs: turning messy criteria into a defensible score.

Bidly (bidly.space): a live SaaS with paying users that reads a job post and generates targeted proposals through the Claude API in about 30 seconds. Same job-post-to-tailored-output pipeline your resume tailoring depends on.

How I'd approach it:
1. Build a parser layer that scores against the real ATS logic (keyword match, section structure, readability) rather than one generic number.
2. Use Claude to extract required skills from the job posting, then rewrite resume sections to match honestly, flagging any claim that isn't backed by the original resume.
3. Return the tailored resume plus a score breakdown so the user sees exactly why they hit or missed 90.
4. Ship it as a Node and React app with export to ATS-safe formats.

Are you tailoring against one specific ATS your applicants keep hitting, or does this need to score well across several different systems?