9 days ago
San Francisco, CA, USAMid Level
Responsibilities
- Design task sets that teach specific coding-agent capabilities and iterate using traces and evaluations.
- Analyze agent traces to identify failure modes and surprising behavior.
- Build reusable systems, rewards, environments, and data-quality processes from one-off solutions.
- Partner with research to evaluate whether datasets teach the intended capabilities.
Requirements
- Strong software engineering fundamentals and the ability to write careful, fast code.
- Ability to break ambiguous capabilities into concrete, measurable tasks.
- Background in infrastructure, data, or distributed systems.
- Interest in analyzing real-world agent behavior and turning it into datasets or tools.
- Reinforcement-learning experience is a plus but not required.
