Halluminate, a nine-person San Francisco startup founded in 2024, has raised $30 million in a Series A led by Oak HC/FT, bringing total funding to $38.5 million. The company builds resettable reinforcement-learning environments where computer-use and browser-based AI agents can practice financial work without affecting production systems. Its offering also includes proprietary benchmarks and expert evaluation services. Halluminate counts four leading U.S. AI labs and the two largest companies developing browser-based AI agents among its customers. The company benchmarks models on financial tasks, identifies weaknesses and converts failure modes into simulated training environments. In an August benchmark, seven frontier models completed 88 tasks based on anonymized private-equity transactions, with the highest average score reaching 51%. Halluminate says the results showed agents struggling to maintain instructions and produce correct final deliverables in complex financial workflows. CEO Jerry Wu said the company has surpassed a mid-eight-figure annualized revenue run rate and is profitable. Halluminate is focusing on frontier model labs rather than broad enterprise expansion, while also contributing to open-source projects including Westworld. Co-founded by Jerry Wu and Wyatt Marshall, both Cornell computer science alumni, the company went through Y Combinator’s Summer 2025 batch.