Research Engineer, Mid-Training
Core
Own late-stage training decisions to sharpen model reasoning, generalization, and capability in coding, math, and long-horizon tasks for AI software agents.
Role type
Senior IC research engineer (mid-training)
Builds
End-to-end software agents (Devin) with advanced reasoning capabilities
Domain
Applied AI / Large Language Model Training
Deliverable
production ML models
Required skills
LLM training pipeline expertise, data mix design, synthetic data pipeline development, optimization theory, distributed training debugging, Python, PyTorch
Preferred skills
Experience with continual pre-training, annealing strategies, context length extension methods, original research contributions
Technologies
PyTorch, Python, distributed training frameworks
Responsibilities
Design and iterate on high-quality data mixtures for late-stage training; Develop synthetic data pipelines for capability improvement; Optimize multi-stage learning rate schedules and compute allocation; Research methods for extending effective context length; Build evaluations to distinguish real capability improvements from overfitting; Measure scaling of training interventions with compute and data