Member of Technical Staff, Frontiers of Deep Learning Scaling

Lead research and engineering efforts to identify and validate scalable paradigms for efficient compute and useful data in next-token prediction.
Design, implement, and run large-scale, long-duration training workflows with robust stability across hundreds of millions of GPU hours.
Build end-to-end pipelines: data preparation, evaluation, experiment design, result analysis, and iterative redesign for faster learning.
Explore novel architectures, learning paradigms (e.g., continual learning, self-improvement), and unified multi-modal models as potential scaling paths.
Collaborate across teams, communicate results concisely, and drive fast, principled experimentation and decision making.
Work with Python, JAX, PyTorch, and Rust to implement ideas, test hypotheses, and deliver measurable improvements.