Halluminate raises $30M to train AI agents for Wall Street work

A startup with fewer employees than a pickup basketball game just closed a $30 million Series A. Halluminate, a nine-person company based in San Francisco, builds training grounds where AI models practice financial work before they ever touch the real thing.

The round was led by Oak HC/FT and lifts the company’s total funding to $38.5 million. For a business founded in 2024, that is a fast climb. The customer list may matter even more: four leading US AI labs, plus the two largest companies building browser-based AI agents.

What Halluminate actually sells

Halluminate creates reinforcement-learning environments for computer-use and browser-based agents. These are AI systems that click, type and navigate software much as a human analyst would.

The sandboxes are realistic and resettable. An agent can make a mess, and the environment snaps back to its starting state for the next attempt. Crucially, none of this disturbs actual production systems.

The company’s offering breaks into three pieces:

• Managed sandbox environments where agents train safely
• Proprietary benchmarks that measure how well models perform financial tasks
• Expert evaluation services that judge agent output

Halluminate also contributes to the open-source community, including a project called Westworld.

The team and the bet

Halluminate was co-founded by Jerry Wu and Wyatt Marshall, both Cornell computer science alumni. Wu previously worked on AI at Capital One Labs. The startup went through Y Combinator’s Summer 2025 batch. Between that program and the new raise, earlier funding accounts for the remaining $8.5 million of its total.

Why the customer list stands out

Halluminate can point to four of the leading AI labs in the US as paying customers. It also counts the two largest browser-agent companies among its clients.

That is a notable roster for a team of nine. It suggests the biggest model developers see value in buying specialized training environments rather than building every one themselves.

What this means

For investors watching the AI sector, the Halluminate round fits a broader appetite for infrastructure plays. These are companies selling tools to model builders rather than competing with them directly.

There are real risks, though. Selling to a small number of very large customers creates concentration exposure. If one major lab decides to build comparable environments in-house, a meaningful slice of revenue could be at stake. Big buyers also tend to have strong negotiating leverage over small vendors.