Han Peng 彭晗
I build multimodal foundation models for coding and agents.
I am a Staff Research Scientist and Tech Lead Manager in Ant Group’s Bailing foundation model group. I built and now lead the multimodal coding and agent post-training team, working across data, training, and evaluation.
Previously, I spent four years at Google. Before that, I was a postdoctoral researcher at Oxford VGG/FMRIB, having earned my PhD in Physics at the University of Oxford.
My broader research interests include multimodal foundation models, reasoning, and learning to act in interactive environments.
Recent Work
Multimodal models · Coding & agents
Ling-3.0-flash-VL
I led the team’s development of the post-training pipeline in multimodal coding, and delivered the agentic coding capabilities in this open-weight vision-language model. With computer use and visual feedback, the model can interact with interfaces, write code, see what it has built, and refine it.
Featured in Hugging Face Daily Papers!
RULER: Instance-aware Rubric Rewards for SVG Generation
Our work on large-scale post-training motivates three practical questions:
- How can we make open-ended tasks meaningfully verifiable?
- How can we align rewards with human taste and aesthetic judgment?
- How can we make feedback detailed and task-specific, yet scalable?
RULER derives evaluation criteria from each prompt, uses a vision-language model to assess the rendered SVG against each criterion, and turns those judgments into reinforcement learning rewards—without paired target SVGs or human preference labels for training.
Background
Previously, I spent four years at Google working on AI for Pixel manufacturing, including automated visual inspection and LLM agents for data analytics. I collaborated with factories, equipment vendors, and solution providers on real-world deployment.
I hold a PhD in Physics from the University of Oxford and was a postdoctoral researcher at Oxford VGG/FMRIB, working on deep learning for brain imaging.
Our team won both tracks of the PAC 2019 brain age prediction challenge. Paper · Results · Code
Professional experience All publications
🚀 We’re hiring!
We are looking for passionate Research Interns and Full-time Researchers to tackle fundamental challenges in AGI. This is an opportunity to work on frontier research and see your work deployed at scale. If you’re excited by our work, please send your CV to: penghan (dot) peng (at) antgroup (dot) com.
For general research enquiry, please reach me at han (at) hanpeng (dot) me.