Klavis AI builds high-quality coding and agentic data for frontier AI post-training. Based in San Francisco and backed by Y Combinator, Klavis works with leading AI labs to create production-grade datasets for RL and SFT training.

We focus on one of the most important bottlenecks in frontier AI: teaching models to reliably write code, use tools, solve complex workflows, and operate in realistic environments. Our datasets combine real-world task design, verifiable rewards, expert rubrics, gold trajectories, and scalable infrastructure for generation and evaluation.

Klavis is founded by engineers with deep experience in agentic AI, ML infrastructure, and large-scale production systems, including prior work on Google Gemini function-calling infrastructure and ML/data infrastructure serving millions of users.