Staff Scientist – Post-Training and Reinforcement Learning for AI for Science
Core
Research and develop post-training methods, including reinforcement learning and preference optimization, to improve the reliability and scientific utility of large-scale foundation models for physics, materials science, chemistry, biology, and climate applications.
Role type
Staff Scientist, hands-on IC research and development
Builds
Scientific foundation models and adaptive learning systems for DOE Genesis mission
Domain
AI for Science / Computational Science / High-Performance Computing
Deliverable
production ML models
Required skills
Reinforcement learning, Post-training methods, Mathematical optimization, Linear algebra, Numerical methods, Python, C/C++, PyTorch, JAX, Distributed training, Large-scale optimization
Preferred skills
Policy optimization, Bandits, Preference learning, Supervised fine-tuning, Direct preference optimization, Reward modeling, Multi-node execution
Technologies
PyTorch, JAX, Python, C, C++
Responsibilities
Develop and scale post-training pipelines for scientific foundation models; Design and evaluate reinforcement learning applications in data-intensive environments; Optimize workflows for leadership-class supercomputers; Partner with domain scientists to apply adaptive learning systems; Conduct original research and publish findings; Address algorithmic and systems challenges in large-scale training.
Seniority
Staff, strategic research & mentorship