Cognition
Research Engineer, Post-Training
San Francisco · On-site · Full-time · Posted Aug 6, 2026
About the job
Develop and evaluate post-training methods that shape Devin’s capabilities, alignment, and behavior in real-world tasks.
What you'll do
- Develop training recipes across datasets, training stages, and hyperparameters, and measure their impact on evaluations and production performance.
- Design and improve evaluations, investigating unexpected training results and distinguishing real effects from noise and bugs.
- Advance alignment and agent-behavior methods, and measure performance scaling with data and compute.
What they ask for
- Demonstrated capability advancing ML systems through post-training, alignment, or related methods.
- Strong foundations in probability, statistics, and ML theory.
- Evidence of original contributions, such as publications at top venues, open-source impact, or equivalent industry results.
- Experience with large-scale distributed training and debugging.
Nice to have
- Experience with RLHF, RLAIF, preference modeling, or reward learning.
- A PhD is one signal among many; the posting does not require it.
Tools
RLHF, RLAIF
AI focus
Post-training, alignment, evaluation, and scaling of AI agents for long-horizon tasks.
Model-written summary · the original posting is authoritative