Jiace Zhao

Hey 👋 I'm Jiace, an incoming undergraduate at UC Berkeley. I'm interested in embodied AI, computer-using agents, reinforcement learning, and robot control. I like building systems that can perceive, reason, and act in the real world.

Tsinghua University High School, I got interested in algorithms and competed in the USA Computing Olympiad, where I reached the Platinum Division. I also started doing independent robotics research with Rujie Wang. I built a 6-DOF robotic arm and developed Adaptive Robust Learning Control to improve the accuracy of low-cost hardware.

Previously, I worked on OSGym at MIT CSAIL, where I was the only high school student in the lab. OSGym is a distributed data engine for training computer-using agents at scale. I worked on task generation, pruning low-quality trajectories, and cleaning training data. I also helped turn these steps into a pipeline that could produce large datasets across many computer environments. This work led to the OSGym paper, which I co-authored.

Outside of research, I like building things people can use. Clingword (Try it ↗) uses reinforcement learning to decide when to introduce new words and when to review old ones. It reached 20,000+ users in under two months. I also built Singularity Academy, where I fine-tuned a GPT model to generate Manim-based educational videos from prompts. It has 2,000+ users and has worked with local schools.

I also produce Singularity Voices, a podcast where I talk with researchers and young founders in China.

I'm currently looking for a research position. I like talking with people about research, ideas, or just life. Feel free to email me at jiace_zhao@berkeley.edu.