Build and improve state-of-the-art agent systems: design RL/RLHF learning pipelines, create custom environments, set up robust evaluation, train smaller task-specific models, leverage open-source text and image models, and help shape technical and hiring strategy alongside founders.