Research Engineer Post Training
CognitionVisit Cognition website
Cognition is an applied AI company that operates Devin, an autonomous software-engineering agent.
San Francisco, United States
Funding history
About Cognition
Cognition builds AI agents and models for software engineering. Its flagship product, Devin, can plan, write, test, and ship code in a customer’s existing codebase and tools.
Skills
Candidate Availability
Required and preferred rules are kept separate and reflect the wording in the original posting.
About the Role
You will develop post-training datasets, training stages, hyperparameters, and evaluation methods. You will investigate unexpected training outcomes, improve alignment and agent behavior with reinforcement-learning approaches, measure scaling with data and compute, and create new methods when current approaches reach limits.
Requirements
- Track record advancing machine learning systems through post-training, alignment, or related methods
- Experience with RLHF, RLAIF, preference modeling, reward learning, or equivalent methods
- Knowledge of probability, statistics, and machine learning theory
- Evidence of original contributions through publications, open-source work, or industry results
- Experience with large-scale distributed training and debugging
- Knowledge of training pipelines, data, and evaluation interactions
Responsibilities
- Iterate on post-training datasets, training stages, and hyperparameters
- Build and improve evaluations for model behavior and production performance
- Investigate and understand unexpected training results
- Apply and advance RLHF, RLAIF, and constitutional alignment approaches
- Measure how performance scales with data and compute
- Develop new methodologies when existing methods reach limits