How Harvey Built a Research Lab on a Budget | Gabe Pereyra

Building a competitive AI research lab as an application-layer company requires leveraging the existing Frontier ecosystem rather than attempting to out-spend major labs. Success hinges on creating domain-specific benchmarks and utilizing synthetic data generation guided by experts, such as lawyers, to navigate sensitive, non-public data environments. Before pursuing post-training, companies must establish rigorous production infrastructure, including automated evaluation, A/B testing, and model routing, to ensure reliability and performance. By open-sourcing datasets, firms can refine their models through community feedback and industry-wide benchmarking. Gabe Pereyra, co-founder and president of Harvey, demonstrates that this strategic approach—focusing on specialized, high-value tasks—allows smaller teams to build frontier-level intelligence and effectively compete in the evolving AI landscape.