Builder and engineer, working where
AI and
robotics meet.
Computer vision Robot learning ML infrastructure Sunnyvale, California Shanghai, China
I have spent the past decade building learning systems that see the world and act in it: sim-to-real RL, computer vision and autonomous-systems perception at Apple, a walking humanoid at Noble Machines, and now visual reasoning models at Elorian.
Member of Technical Staff · Robotics
Elorian is a visual reasoning lab founded by former Google DeepMind researchers, building native multimodal models that reason directly over images, video, audio and text rather than translating everything into words first. I lead the robotics vertical, bringing that kind of visual reasoning into embodied settings.
Co-founder, now Advisor
Humanoid robots for the hazardous, physically demanding industrial work that keeps the world moving. We started as Under Control Robotics out of South Park Commons, where a team of four built a walking humanoid prototype from scratch in about four months. I have since stepped back into an advisory role.
Staff Machine Learning Engineer
Eight years, from junior engineer to Staff. Computer vision first — including SimGAN, a sim-to-real approach that won the CVPR 2017 Best Paper award — then perception for autonomous systems in the Special Projects Group, with stretches in NLP, recommendation systems, and the distributed training infrastructure underneath all of it.
M.S. in Robotic Systems Development (MRSD) · Robotics Institute
Perception and autonomy, back when self-driving cars were the problem everyone wanted to work on.
Language models and world models have each succeeded on their own terms. AGI will come from unifying them correctly, not from choosing between them.
If the LLM is a compiler, the natural-language blueprint is the source of truth — and “vibe coding” has to grow up into intent-based engineering.
CVPR 2017 · Oral · Best Paper Award
SimGAN: refine synthetic images with unlabeled real data so a simulator's free labels transfer to the real world. State of the art on gaze and hand-pose estimation without a single labeled real image.
With CMU's Robotics Institute — on eight years at Apple, South Park Commons, and why humanoids for heavy industry felt like the frontier worth leaving for.
The Robot Report on the company we started — a rugged humanoid deployed with a Fortune Global 500 customer within eighteen months.