Research Engineer, Mid-Training

The role

You will own late-stage training decisions including data mix, annealing schedules, and context length extension. This role involves research and engineering to sharpen model capabilities in coding, math, and reasoning.

What you will do

• Design and iterate on high-quality data mixtures for late-stage and annealing training runs.
• Drive targeted improvements in coding, mathematics, and long-horizon reasoning.
• Develop and evaluate synthetic data pipelines.
• Research and optimize multi-stage learning rate schedules and compute allocation.
• Research and implement methods for extending effective context length.
• Build evals to distinguish real capability improvements from benchmark overfitting.
• Measure how mid-training interventions scale with compute and data.

Required qualifications

• Deep familiarity with the LLM training pipeline end to end
• Hands-on experience with continual pre-training, annealing, or late-stage data mixing for large models
• Strong intuition for data quality
• Experience developing or evaluating synthetic data pipelines for capability improvement
• Proficiency in Python and deep learning frameworks (PyTorch)
• Comfortable debugging distributed training at scale
• Strong fundamentals in optimization, statistics, and ML theory
• A track record of original contributions: publications, open-source impact, or internal results that moved a capability frontier
• Comfort operating in ambiguous, fast-moving environments

Nice to have

• A PhD is one signal among many

Benefits

• Large compute allocations with training jobs running across thousands of GPUs

About the company

Cognition is an applied AI lab building end-to-end software agents. They are the makers of Devin, the first AI software engineer.