Software Engineer, RL Training Infra

The role

This role supports OpenAI's Post-Training Frontiers team, which develops reinforcement learning systems for frontier agents deployed across Codex, ChatGPT, and the API. The engineer acts as a generalist troubleshooter, tackling urgent infrastructure problems that emerge during large-scale RL training runs, from distributed system failures to inference bottlenecks and scaling challenges.

What you'd do

• Unblock large-scale RL training runs by addressing urgent engineering and infrastructure bottlenecks
• Debug issues spanning training systems, inference, orchestration, scaling, and distributed infrastructure
• Solve technical problems at the intersection of research and engineering, including scaling experiments and reliability improvements
• Support researchers building infrastructure-heavy features like multi-agent capabilities and memory systems
• Convert recurring operational issues into improved tools, systems, and abstractions
• Collaborate with research, infrastructure, and partner teams during critical model training timelines
• Debug failures across model behavior, training data, RL systems, evaluation infrastructure, and serving systems

What they're looking for

• Strong generalist engineer with experience in at least one layer of ML infrastructure
• Background in RL, inference, scaling, training systems, orchestration, or similar ML infrastructure domains
• Ability to learn rapidly and operate across unfamiliar technical layers
• Strong debugging skills with high ownership and excellent communication
• Comfort operating in ambiguous, fast-moving environments with tight timelines
• Motivated by reliability, speed, and building load-bearing systems

Nice to have

• Experience supporting large-scale model training or async RL systems
• Background debugging distributed systems across GPUs, networking, orchestration, or inference
• Performance optimization or production-critical infrastructure expertise
• Direct experience working with research teams or fast-moving model development groups