AI · Article or course
People Search Bench
Open benchmark for evaluating AI-powered people search agents and platforms
People Search Bench is an open evaluation framework and dataset designed to test how accurately AI agents find real people from natural language queries. Rather than relying on subjective LLM-as-a-judge scores, it decomposes queries into verifiable criteria and validates candidates against live web sources across 119 diverse tasks.
What it's for
- Evaluate the accuracy and coverage of AI agents in finding professional contacts and leads
- Benchmark search engines and agent tools across recruiting, B2B prospecting, and expert discovery
- Stress-test retrieval workflows on multi-constraint people-search queries
Features
- 119 benchmark queries spanning four categories: Recruiting, B2B Prospecting, Expert/Deterministic, and Influencer discovery
- Multilingual coverage featuring queries in English, Spanish, Portuguese, and Dutch
- Criteria-grounded scoring pipeline using Tavily web search to verify returned candidates against factual requirements
- Three standardized metrics: Relevance Precision (padded nDCG@10), Effective Coverage, and Information Utility
- Head-to-head comparisons across tools like Lessie, Exa, Juicebox, and Claude Code
Pricing
Open source. Free and open source (MIT license)
Details
- Type
- Article or course
- Category
- AI
- Topic
- Model Benchmarks
- Open source
- Yes
- Platforms
- Code library
- For
- AI researchers, agent developers, and B2B data tooling engineers
- Alternative to
- MT-Bench
- Tags
- benchmark, ai agents, people search, lead generation, evaluation, prospecting
Why it was saved
- @jasonzhou1993: Independent-looking benchmark the results are measured against; Claude Code alone 43% vs Claude Code + treg 78.2%.
Related
No-code visual canvas for building AI mini-apps from plain English prompts
Open-source tutorials and project templates for LLMs, RAG, and AI agents.
Official repository for Anthropic Agent Skills and Claude Code plugin templates
AI no-code platform for building full-stack web applications and websites.
Open-source, self-hosted AI agent with persistent memory and messaging integrations.
Design inspiration MCP server and archive for AI coding agents.
Updated