Dave Van Veen
AI Researcher | Google DeepMind | Stanford PhD
I build AI systems people can trust.
Currently, I'm a research scientist at Google DeepMind on Gemini. My work spans post-training and agentic tool use, with a focus on factuality and grounding.
I write about agentic systems and AI's societal impacts. Away from the computer, you'll find me reading, surfing, or trail running with my dog.
bio
Previously, Dave was a principal AI scientist at a start-up, building the vision-language models behind its first product. He earned his PhD from Stanford under Akshay Chaudhari and John Pauly, developing algorithms for computational imaging and multimodal models. He has published 15+ papers (3000+ citations) across Nature, Nature Medicine, CVPR, ACL, et al.
Prior to his PhD, Dave was a research scientist at a Bay Area start-up and Stanford's AIMI Center. He earned his MS at the University of Texas doing ML research under Alex Dimakis and Sriram Vishwanath, during which he served as a Data Science for Social Good fellow in London. As an undergrad at Wisconsin, he founded Badgerloop, a 150-person team that built a hyperloop pod for SpaceX.
news
| Jun 2026 | Integrated Meta’s V-JEPA 2.1 into Hugging Face Transformers. |
| Apr 2026 | New post! Designing agentic systems to scale judgment across AI workflows. |
| Apr 2026 | Shipped Hugging Face TRL in RadEval, enabling direct optimization of radiology metrics via RL. |
| Apr 2026 | Introduced cognitive convergence: what happens when AI shapes human thought at scale. |
| Mar 2026 | Merlin, our 3D CT foundation model (170K+ downloads), published in Nature. |