Sebastian Crossa

Sebastian Crossa is cofounder of ZeroEval, a New York City company helping teams evaluate and improve AI agents with judges that learn from labeled mistakes. Crossa and cofounder Jonathan Chavez also built LLM Stats, a public LLM leaderboard and independent evaluations site. They joined the YC Summer 2025 batch.

Mexico City friends turned founding engineers

Crossa and Chavez met in their first year of college in Mexico and spent years shipping side projects together. Crossa was a founding engineer at Micro (email, a16z-backed) and at Atrato Pago (YC W21). Chavez was an early employee on Datadog's LLM observability team and did undergrad research on vision transformers and robotics RL. In high school Crossa also built and scaled Minecraft servers in his spare time.

From public leaderboards to self-improving judges

LLM Stats grew into a widely used archive of benchmark results and independent evals aimed at contamination-proof measurement. ZeroEval extends that eval instinct into product form: calibrated LLM judges that improve as teams label failures on multi-turn agents, plus Autotune-style sweeps across models and prompts from a small set of human samples. Their Launch YC note puts the bar plainly - your agents are only as good as your evals.