Halluminate
Benchmarks and simulated workplace environments that help AI labs train and evaluate agents on financial analysis and other professional knowledge work.
Product
Teaching agents through complete work assignments
Halluminate builds simulated workplaces where AI agents complete professional assignments using documents, spreadsheets, a data room, email and chat. Finance experts construct the tasks and scoring criteria, turning the quality of an agent’s deliverable into a signal for reinforcement learning and evaluation. The aim is to train models on extended workflows, such as reconciling deal information and producing a financial model, that are difficult to learn from static text alone.
Its customers are AI model labs. Oak HC/FT says frontier labs already use the environments to train financial and economic reasoning. Halluminate also publishes research: its Westworld Finance Diligence Bench contains 88 tasks based on anonymized private transactions, covering modelling, earnings analysis, diligence, presentations and closing. This provides a concrete evaluation of complete deal work; the lab is expanding from financial services and consulting into additional professional fields.
Funding history
1 round- $30M