CueBench YC 申请

CueBench

RL post-training for science reasoning & engineering

CueBench is post-training reinforcement learning for science reasoning and performance engineering. We help data engineering and ML teams use RL to specify, optimize, and audit complex reasoning and compute pipelines - especially those involving scientific or technical tasks. The product automates fine-tuning on real usage data, with a focus on hard-to-specify objectives and reliability for B2B. Our main users are teams who care about model reliability and performance when deploying ML in production.

We all live in San Francisco, CA, and that's where the company will be based after YC.

We're in San Francisco because it's the best place for technical B2B companies - talent, early customers, and peers are all here. Being local also helps us work closely as a small engineering team.