PSB-Recruiting benchmark
Tavi scores 97.03 on recruiting search.
Tavi was evaluated on PSB-Recruiting, the 30-query recruiting category of PeopleSearchBench. The benchmark measures whether a people-search system finds relevant candidates, returns enough qualified profiles, and provides useful information for review.
- Overall
- 97.03
- Relevance
- 97.26
- Coverage
- 99.90
- Utility
- 93.94
The best people search in the world.
Tavi scores 97.03 on PSB-Recruiting, 3.6 points ahead of the next-best system.
Overall score on PSB-Recruiting, 0–100. Higher is better.
Methodology
How the benchmark works
Each recruiting brief is decomposed into explicit criteria. Returned candidates are checked against those criteria using web evidence and an LLM judge, then scored for ranking quality, candidate coverage, and information utility. The underlying benchmark paper is available on arXiv.
- The test uses 30 recruiting searches, each written like a real hiring brief.
- For each brief, Tavi can return up to 15 people, ranked from strongest fit downward.
- Each returned person is converted into the same structured format the benchmark expects.
- The evaluator breaks every hiring brief into explicit criteria, then checks each candidate against them.
- Web search is used to gather outside evidence, so judgments are grounded in public information.
- An LLM judge reviews the evidence and scores relevance, coverage, and information usefulness.
- The evaluator is run 7 times, then the results are averaged to reduce judge variance.