TrainLoop

Jackson Stokes spent years inside Google optimizing Gemini and AlphaFold—the same models now powering ChatGPT-chasing products across Silicon Valley. Mason Pierce and Isheeth were right there with him. Now they're building TrainLoop, and they're solving the problem nobody else is talking about: reasoning models are incredible on benchmarks, but your business isn't a benchmark. Your business is codegen, compliance, legal documents, healthcare—vertical-specific pain points that generic AI can't touch.

The big players all have internal fine-tuning pipelines. Google does. OpenAI does. The rest of us? We're stuck with whatever pretrained models decide to give us. TrainLoop is bringing that same enterprise-grade reasoning fine-tuning capability to every developer building AI into production. They're in Y Combinator now, and if the vertical AI wave is really the next big thing—which it absolutely is—then whoever owns the fine-tuning layer owns the stack. These founders already built the best reasoning models in the world at Google. Now they're letting the rest of us in on the secret.