Research Engineer, Post-Training

About this role

Join an applied AI lab as a Research Engineer focused on post-training systems for Devin, an AI software engineer. You'll develop training recipes, design evaluations, and advance alignment techniques that determine how AI agents learn and behave in real-world tasks.

What you'll do

• Iterate on post-training datasets, training stages, and hyperparameters to optimize model behavior across evaluations and production
• Design and maintain evaluation frameworks that capture meaningful performance metrics and catch capability gaps
• Investigate unexpected training results to understand root causes and apply insights to future improvements
• Implement and advance alignment techniques including RLHF, RLAIF, and constitutional approaches for long-horizon agent tasks
• Analyze scaling relationships between data, compute, and performance; develop new methodologies when existing approaches plateau
• Bridge research and engineering by moving from prototypes to deployed systems with minimal process overhead

What they're looking for

• Post-training methods (RLHF, RLAIF, preference modeling, reward learning)
• Probability, statistics, and machine learning theory
• Large-scale distributed training and debugging
• Experimental design and statistical analysis for distinguishing signal from noise
• Systems-level thinking across training pipelines, data, and evaluation
• Python and deep learning frameworks
• Ability to work in ambiguous, fast-moving research environments

Benefits

• Frontier-scale compute resources with GPU allocations available from day one
• Small, highly selective team where research directly impacts product deployment
• Minimal bureaucracy with emphasis on speed and technical autonomy
• Work on one of the most competitive problems in AI