Frontier foundation models · Pretraining · Post-training · RL
I build learning systems that move from research ideas to frontier-scale training and real products.
Research scientist with 15+ years building frontier foundation models and large-scale learning systems. I work across the full model lifecycle—from data and scaling ladders through pretraining, mid-training, SFT, reinforcement learning, evaluation, and deployment. Contributor to Grok, Gemini, and Apple Vision Pro, with particular depth in multimodal learning, generative modeling, and large-scale RL.
Selected work
My work connects frontier-model training, reinforcement learning, generative modeling, and spatial intelligence.
xAI · Frontier Models
Frontier-model development across scaling ladders, pretraining, mid-training, post-training, evaluation, large-scale data, and successive Grok releases.
Gemini
Google · Foundation Models
Billion-scale data construction, end-to-end training and controlled ablations, model adaptation, and release evaluation integrated with Gemini core training pipelines.
Meta FAIR · Reinforcement Learning
A large-scale open reproduction and analysis of AlphaZero, spanning distributed self-play, training, inference, and evaluation. ICML 2019 Oral.
Meta FAIR · Generative Modeling
Order-aware autoregressive generation for structured 3D worlds—an early bridge between generative modeling and spatial intelligence.
Vision Pro
Apple · Spatial Intelligence
Led production 3D perception with voxelized RGB-D inputs, temporal fusion, detection, and scene reconstruction deployed in Apple Vision Pro.
Experience
Member of Technical Staff
Omni / Multimodal Foundation Models
Staff Research Engineer & Tech Lead
Vertex AI / Gemini
Staff Research Scientist & Tech Lead
Applied Foundation Models
Senior Research Scientist
Autonomous Driving Research
Research Engineer
Reinforcement Learning & Generative Models
Senior Research Scientist
Institute of Deep LearningEducation
Publications & technical releases
Selected model releases, conference papers, technical reports, and foundational work from 2009–2026. For citation details, see Google Scholar ↗.