Score job candidates to build a first-pass shortlist
Jev scores every candidate against a role so a smaller shortlist goes to closer review and, finally, the hiring manager.
Author's figure: TypeSafe reports screening cost per 1,000 candidates fell from $0.755 to $0.092 (88%) with shortlist recall roughly unchanged in an offline test on 150 roles.. Reported by the builder, not measured by us.
The use case
Recruiting teams and talent marketplaces can face thousands of candidates per open role, too many for close review. Jev scores each candidate profile against the role and the top slice becomes the shortlist. A stronger model or a recruiter then reviews that shortlist in depth. In the published case, Jev replaced a small LLM for this first scoring stage. The hiring manager still chooses who to meet, and a person should audit the cut-off regularly for missed strong candidates and for unfair patterns, since this is an employment decision.
Example workflow
- Pull candidate profiles for an open role from the ATS or marketplace database
- Ask Jev to score each profile's fit against the role requirements
- Keep the top-scoring candidates as the shortlist
- Send the shortlist to a deeper model review or a recruiter
- Hiring manager picks who to interview; audit the cut-off against past hires
Tools
- Jev (TypeSafe AI)
- Applicant tracking system (e.g. Greenhouse or Lever)
- Python
Risks and guardrails
- Strong candidates can fall below the cut-off without anyone noticing
- Scoring can encode bias; hiring rules may require audits and notices
- Profiles with thin text may score unfairly low
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- TypeSafe AI customer story: Jack & Jill: typesafe.ai/blog/jack-jill-jev-case-study
- Hacker News submission: news.ycombinator.com/item
- GitHub repo gtaras7/typesafe-jev (cv-screen workbench with editable scoring policy): github.com/gtaras7/typesafe-jev