I’m a Generative AI Software Engineer at ByteDance. I build systems that bring generative and multimodal models into creative tools: image and video generation, content understanding, long-running jobs, and the infrastructure behind them.
My path into this work started with computer graphics and 3D vision. I received an M.S. in Computer Science from the University of Southern California in 2023, after a B.E. in Computer Science and Engineering from the Chinese University of Hong Kong, Shenzhen. Along the way, I worked on generative models at TikTok/ByteDance and Ant/Alibaba, and on single-view 3D reconstruction at the Shenzhen Research Institute of Big Data.
These days I am especially interested in the unglamorous part of applied AI: making a promising model dependable enough to become a product. That usually means careful interfaces, recoverable workflows, useful evaluation, and enough observability to understand what happened when things go wrong.
This site began as a place for notes on graphics and deep learning. Those notes are still here. I am also using it to write about the engineering ideas that have stayed useful as my work has moved from research prototypes to creative AI systems.
Current interests
- Generative and multimodal AI systems
- Creative tools, graphics, and human-computer interaction
- Reliable inference services and workflow orchestration
- Tools that make AI capabilities easier to compose and evaluate
Selected work
One-Shot Generative Domain Adaptation (GenDA)
GenDA studies how a GAN pre-trained on one image domain can be adapted to a new domain from as little as one target image. This work was completed during my TikTok/ByteDance internship with Dr. Yujun Shen and Ceyuan Yang.
VIPNet: Single-view 3D reconstruction
VIPNet is a sparse implicit point-guided approach to reconstructing high-resolution 3D shapes from a single image. It was published at 3DV 2020.
C++ 3D Graphics Renderer
I built a small OpenGL-like renderer to understand the full path from a 3D model to an image: rasterization, transformations, shading, texturing, and anti-aliasing.
Personal creative work
Outside engineering, I also enjoy illustration and visual identity design.
Breakfast poster
I made this poster for a campus breakfast activity. Its Chinese title, “早点来吧”, is a small pun: it can mean both “come early” and “come have breakfast.”
STAr logo
This logo was made for Sailboat Test Arena (STAr), a project for testing and guiding autonomous sailboats. The Big Dipper is a nod to both navigation and the project’s name.