HireScope

Problem

Most resumes are read by software before a person sees them. An applicant tracking system, an ATS, parses the file and matches it against the job description. A good candidate with a two-column PDF can lose before anyone reads a word.

Approach

spaCy handles the parsing and entity extraction. sentence-transformers turn both the resume and the job description into vectors so they can be compared by meaning rather than by exact words. A RAG pipeline, retrieval-augmented generation, pulls the relevant parts of the job description before generating feedback, so the advice points at the actual posting.

FastAPI serves it, React renders it.

Results

87% ATS pass rate on the resumes tested.

Limits, and what I would do next

The pass rate is against the parsing rules I modelled. Every real ATS is a little different and none of them publish their rules, so this predicts a class of behaviour rather than one specific product.

Optimising for the filter is a narrow goal. The genuinely useful version would tell you when your resume is fine and the job is a bad fit, which no tool has an incentive to say.