Agentic Job-Application Pipeline
A queue-driven agent that scrapes, scores and applies to job postings end to end, tailoring a résumé per posting without letting the model invent experience.
Applying to jobs properly means reading each posting and rewriting a résumé to match it — perhaps 40 minutes each. Skipping that and mass-applying with one generic CV is what everyone does, and it is why nobody hears back.
A queue-driven pipeline handling up to ~6,900 postings a day, cutting hands-on time per application by ~90% while a declared-profile constraint keeps the generated résumés truthful.
About this project
This pipeline handles the full application loop: it collects postings, deduplicates them by unique constraint, scores each one for fit, and generates a résumé tailored to the posting's own vocabulary. The interesting constraint is truthfulness. A deterministic mini-ATS scores each posting before any model is involved, so obviously off-stack roles are filtered without spending a token. When the model does tailor a résumé, it is given exactly one declared profile file as the only facts it may use, with hard rules against introducing a technology that isn't there or claiming seniority the profile doesn't support — the generator then cross-checks its own output against that file for drift. The result is a queue-driven system handling up to ~6,900 postings a day behind a REST API, a PostgreSQL store and a Next.js dashboard for reviewing and editing drafts, cutting hands-on time per application by roughly 90%.
Key Features
Tech Stack
