How Harvey Built a Research Lab on a Budget | Gabe Pereyra
Harvey built a domain-specific research lab by leveraging frontier ecosystem rather than competing with it directly.
“it's an unfair game competing with the frontier labs if you're an application layer company.”
Harvey co-founder Gabe Pereyra outlines the company's playbook for building a research lab at the application layer: start with domain-specific benchmarks, use them to drive post-training via frontier model partnerships, then serve customized models in production. Harvey released three legal AI datasets this year—Legal Agent Bench, a contracting dataset, and a diligence dataset described as one of the largest RL environments ever released, with data rooms up to 80 million tokens. The talk serves as a template for how vertical AI companies can build frontier-quality intelligence without frontier-scale resources.