Research Engineer – Post‑Training

About the role

cognition is an applied AI lab building end‑to‑end software agents, including Devin, the first AI software engineer. The Post‑Training Research Engineer will bridge raw model capability and real‑world usefulness, safety, and effectiveness.

Key responsibilities

• Iterate on full‑stack training recipes, including datasets, stages, and hyperparameters.
• Design and maintain evaluation pipelines that capture meaningful performance metrics.
• Diagnose unexpected training outcomes and translate insights into future improvements.
• Apply and advance alignment techniques such as RLHF, RLAIF, and constitutional AI.
• Measure scaling behavior with data and compute, proposing new methods when limits are reached.

Required profile

• Proven track record advancing ML systems through post‑training, alignment, or related methods.
• Strong fundamentals in probability, statistics, and machine‑learning theory.
• Evidence of original contributions (publications, open‑source impact, or industry results).
• Experience with large‑scale distributed training and associated debugging.
• Systems‑level thinking about pipelines, data, and evaluation.
• Comfort with ambiguity and fast‑moving research environments.
• PhD is a signal but not mandatory.

Required skills

• RLHF
• RLAIF
• Preference modeling
• Reward learning
• Large‑scale distributed training
• Probability
• Statistics
• Machine‑learning theory