Software Engineer, RL Training Infra

The role

This role sits within the Post-Training Frontiers group, which trains the agentic models behind several OpenAI products. The engineer works across the full stack of a large RL training effort, stepping into whichever reliability, scaling, or systems problem is most pressing, and also helps researchers land infrastructure-heavy capability work.

What you would do

• Respond to the most urgent engineering and infrastructure failures in large RL training runs
• Track down root causes across training, inference, orchestration, and distributed systems
• Tackle scaling, latency, and cost challenges at the seam between research and engineering
• Improve uptime and throughput of RL training jobs
• Support researchers building infrastructure-intensive features like multi-agent behavior or memory
• Convert repeated operational pain points into reusable tooling and processes
• Coordinate with research, infrastructure, and partner teams during compressed model release schedules
• Investigate failures spanning model behavior, data, evaluation, serving, and agent harnesses

What they asked for

• Broad software engineering background with exposure to at least one layer of ML infrastructure
• Experience with RL, inference, training systems, orchestration, or related ML tooling
• Strong debugging instincts and a track record of owning messy problems end to end
• Fast learner who can ramp up in unfamiliar parts of the stack
• Clear written and verbal communication and a collaborative, unpretentious working style
• Comfort with practical, unglamorous reliability work when the team needs it

Nice to have

• Hands-on work with large-scale model training, asynchronous RL, or high-throughput ML systems
• Debugging experience across GPU clusters, networking, scheduling, or inference stacks
• Performance tuning or production-critical systems background
• Day-to-day collaboration with research scientists or fast-paced model development groups