Research → Product
Models shipped globally in Windows Copilot Vision, PowerPoint Designer, and Microsoft Designer.
Miaosen (Jackson) Zhang
I build computer-use agents and multimodal reasoning systems— from frontier research to models used in Windows Copilot and Microsoft Designer.
MODEL IN PRODUCTIONWindows Copilot Vision Highlights
GLOBAL
My work connects model capability, scalable data, and real product impact. I focus on the missing layer between language models and useful action: perception, grounding, and reasoning in complex environments.
Models shipped globally in Windows Copilot Vision, PowerPoint Designer, and Microsoft Designer.
Efficient 3B–4B multimodal systems that compete with substantially larger models.
Core contributions to LMM-R1, XwinLM, OpenRLHF-M, and reusable model-training infrastructure.
Select a project to enter the details. The collection is designed to grow with the next breakthrough.
From screen perception to the full human action space.
Five years of building models, datasets, and systems across the full research-to-production loop.
Building computer-use agents as software and UX testers with the MAI Copilot team.
Led data generation and GUI grounding model training for the globally shipped Vision Highlights experience.
Built visual retrieval and aesthetic alignment models deployed across Designer experiences.
B.S. in Computer Science; joint Ph.D. with Microsoft Research Asia, advised by Baining Guo.
Research. Engineering. Impact.
I’m interested in ambitious teams working on agents, multimodal intelligence, and post-training.
zhangmiaosen0215@gmail.com ↗