// the find
vasu-devs/JustHireMe
Local-first AI job intelligence workbench for scraping roles, ranking fit, and generating tailored application materials.
JustHireMe is a local-first desktop app for job search automation: scrape listings from various sources, run them through a quality gate, score fit against your profile using graph + vector matching, and generate tailored resume/cover letter PDFs. It runs entirely on your machine — no cloud account needed, data stays local. Aimed at people who are done with noisy job boards and black-box apply tools.
The architecture separates concerns properly: a deterministic quality gate rejects junk before anything else touches it, then a scoring engine combines source signal, seniority fit, stack coverage, and semantic similarity with visible reasons for each score — not a single opaque number. The Kuzu graph DB for profile relationships alongside LanceDB for vector search is a thoughtful choice; relational queries on your career history plus semantic lookup without either being shoehorned into the other's data model. Local ONNX embeddings (all-MiniLM-L6-v2) mean core matching works offline with no API key. The test suite is serious — 80+ files covering unit, integration, regression, API auth, and e2e — for an open-source tool at this stage that's a signal someone cared about correctness.
The stack is a genuine contributor barrier: Tauri (Rust) + React + Python 3.13 + SQLite + Kuzu + LanceDB + optional Playwright means five different runtime environments before you can run the full app; the quick-start alone requires Node 24, Python 3.13, Rust stable, and uv. The 'thin installer with first-run download' story is fragile — corporate networks, proxies, and slow connections will break it in ways that are hard to debug for non-developer users, and job seekers aren't always developers. `all-MiniLM-L6-v2` is 384-dimensional and trained on general sentence pairs, not job descriptions; the claim that semantic matching works well across healthcare, trades, finance, and software is optimistic — this model will produce noisy similarity scores between job postings and profile text in niche domains. macOS is not notarized, which means 'the app is damaged' warnings on first launch for the platform where a lot of developers live; it's on the roadmap but it's real friction today.