AI agent training and simulation infrastructure
Theta Software
Theta Software builds simulation environments that train AI agents for specialized business work. It serves enterprises and frontier AI labs, combining customers’ first-party data with human expert data to recreate workflows, tools and business goals. Its custom environments support tasks such as financial modeling, invoice processing and code refactoring. Theta also works directly with customer teams to define success criteria, build training environments with feedback and observability, and deploy agents into their operations.
Founded in 2025, Theta’s cofounders are Rayan Garg, CEO; Tanmay Sharma, CPO; and Gurvir Singh, CTO. Its initial memory layer analyzes previous agent runs for critical steps and mistakes, then retrieves relevant insights to plan subsequent runs. Its CUB benchmark extends this focus on learning from work to evaluation: 106 end-to-end computer and browser workflows across seven industries, developed with domain experts. The accompanying infrastructure supports browser and desktop environments, using parallelization and virtual-machine snapshots for evaluation and post-training.
Theta participated in Y Combinator’s Spring 2025 batch. Its investment partnership with Peak XV began in 2025, and Theta is a seed-stage company.
1 talk
Newest first
1 speaker at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.
Messages from the stage
Evaluating outcomes and execution paths
The talk describes judge or critic models that assess both final environment state and agent trajectories, alongside queryable execution traces for examining agent behavior.
Affiliations reflect each recorded session, not necessarily current employment.