Halluminate
Category: AI Infrastructure
Data research lab that builds benchmarks and reinforcement-learning environments to train frontier AI agents on complex knowledge work, starting with financial services. Halluminate was founded in 2024. The company is led by Jerry Wu. Based in San Francisco, United States. Team size: 1-10. Total funding raised: $38.5M. Latest round: Series A. Key investors include Oak HC/FT, Y Combinator, Orange Collective, Heavybit, Antigravity Capital, FT Partners.
- Founded
- 2024
- Headquarters
- San Francisco, United States
- Team size
- 1-10
- Total funding
- $38.5M
Value proposition
Halluminate benchmarks frontier models on realistic, long-horizon financial workflows such as due diligence, modeling and deliverables. It turns the failure modes it finds into high-complexity RL training environments and data for post-training.
Products and solutions
RL environments: simulated desktop, computer-use and tool-use sandboxes modeled on real software and data rooms. Proprietary benchmarks: Westworld Finance Diligence Bench (88 PE due-diligence tasks), Westworld simulated-internet benchmark, and BrowserBench. Expert-annotated evaluations and error analysis. Domain expert network.
Unique value
A verticalized data research lab. It specializes deeply in finance, using practicing deal professionals and domain-specific verifiers, and keeps doubling environment complexity (its 'Moore's law of environments') every 6-8 months so environments stay useful for training frontier models.
Target customer
Frontier foundation model labs, especially leading closed-source US AI labs. Enterprises are a later, secondary target.
Industries served
AI/foundation model labs, financial services (investment banking, private equity), consulting, expanding to insurance, operations and other knowledge-work verticals.
Technology advantage
Finance domain expertise and an expert network. Benchmarks built from anonymized real private transactions. Mixed agentic, discrete and binary verification pipelines. Long-horizon environments with hundreds of steps. Profitable at a mid-eight-figure annualized revenue run rate with a nine-person team.
How they differentiate
Narrow vertical specialization in finance and knowledge work rather than horizontal data labeling. Benchmark-first methodology that converts measured agent failure modes into RL environments. Deliberate focus on a small set of frontier labs.
Main competitors
Mercor (incl. acquired Deeptune), Scale AI, Surge AI. Also RL-environment startups such as Mechanize and Fleet AI.
Key partnerships
Y Combinator (Summer 2025 batch). Paying customers include four of the five leading closed-source US AI labs (unnamed).
Notable customers
Four of the five top closed-source US AI labs (unnamed)
Major milestones
2024: founded by Jerry Wu and Wyatt Marshall. 2025: Y Combinator Summer 2025 batch, launched computer-use sandboxes, the Westworld benchmark and BrowserBench, prior funding of about $8.5M implied. Aug 2026: released Westworld Finance Diligence Bench. Oct 2026: $30M Series A led by Oak HC/FT, bringing total funding to $38.5M.
Growth metrics
Grew from $0 to a mid-eight-figure annualized revenue run rate in about 10 months and is profitable. Nine employees as of October 2026, with plans to grow.
Market positioning
Specialist, profitable provider of post-training RL environments and benchmarks for frontier labs, in the fast-growing RL-environment market.
Geographic focus
United States (frontier AI labs)
About Jerry Wu
Ex-Capital One Labs Product & Research Lead (launched one of the first AI agents in financial services, co-authored 3 patents); Cornell University BA Computer Science & Economics, researched model quantization; VP Cornell Consulting Group.
Latest news about Halluminate
More AI Infrastructure companies
Official website: https://halluminate.ai