Review AI training problems and research environments to assess realism, design quality, and grading accuracy (do not solve). Provide constructive feedback and detailed annotations for 30–90 minute tasks in a remote, contractor capacity.

Design and build reinforcement-learning training environments and diverse tasks to evaluate and improve LLM agents; iterate rapidly on task designs from customer feedback, deliver high-quality outputs with minimal supervision, and maintain PST overlap for collaboration.