Halluminate

Halluminate is a data research lab, founded in 2024 and based in San Francisco, that builds the training substrate on which capable AI agents are made: reinforcement-learning environments, benchmarks, simulated computer-use and tool-use environments, and high-quality expert data. Its stated aim is realistic training environments for complex knowledge work, and its mission is the advancement of AI agents capable of reliable, economically useful computer work. The lab serves AI labs and enterprises.

The work spans reinforcement-learning environments, AI benchmarks, computer-use simulation, tool-use environments, expert data curation, reward systems, model research and AI agent evaluation. The team combines domain practitioners with engineers, environment designers, model researchers and reward-system specialists, and was founded by Jerry Wu and Wyatt Marshall.

Its flagship release, Westworld Finance Diligence Bench, evaluates AI agents across 88 problems drawn from anonymized private transactions and a complete company-acquisition due-diligence process. Alongside it, Halluminate operates an expert network, working with financial professionals to convert real workflows and institutional knowledge into training environments. The verticals served include financial services, consulting, investment banking, private equity, financial analysis and due diligence, with expansion under way into adjacent knowledge-work verticals.