I'm Boyao Xu, co-founder of Benchology. We benchmark B2B data providers: which one actually finds the right email, phone number or company record for the people a team sells to. I build our benchmark infrastructure, and the API and MCP server we're building so GTM engineers can run those tests themselves. Our public studies compare seven web search APIs on 7,000 queries and Apollo against Clay on 2,406 people across five ICPs, with their data files published under CC BY 4.0. Before Benchology I did quantitative trading research and built tools that read Reddit and event data to spot demand. I study economics and mathematics at Université de Montréal and live in San Francisco.
Now
Updated- Preparing pilots of the Benchology API and MCP server with GTM engineers.
- Preparing the December re-runs of both public benchmarks.
Selected work
Web search API benchmark. Seven APIs on the same 1,000 SimpleQA questions, measured on retrieval accuracy, latency and cost.
GTM data benchmark: Apollo vs Clay. 2,406 people, five ICPs, rows checked against LinkedIn, results by segment instead of one overall score.
- –
Quantitative trading research. An event-driven strategy on delayed repricing of small caps after shocks to connected large caps, using a supply-chain graph built from filings and earnings calls. Backtest only: 1.7 net Sharpe at 10% volatility.
- –
SeedCast Prediction. Daily Reddit and Hacker News ingestion; recurring complaints clustered and tracked week over week. Source on GitHub.
- –
SeedCast Event. 27,000+ San Francisco events, enriched to help B2B teams find prospects in person.
- –
TongWenGuan. An AI English-learning app; 200 beta testers in 48 hours. It taught me not to over-engineer before distribution.