Real-time audio-video understanding
Algorithms for live and recorded media understanding with strict requirements on latency, throughput, and robustness.
I work as a researcher and algorithm engineer with a focus on large-scale real-time audio-video understanding. Over the past few years, my work has centered on video and livestream products with demanding real-time requirements.
I am most interested in problems that need both algorithm depth and practical impact: work that requires strong modeling intuition, careful evaluation, and technical direction that still holds up at real scale.
What I Work On
The common thread across my work is studying and building algorithms that are expressive enough for rich media and robust enough for large-scale production environments.
Livestream and video intelligence
Research and development for large-scale video and livestream intelligence.
Generative system design
Part of my work explores generative directions where generation needs to coexist with understanding tasks and practical constraints.
Algorithm direction
Setting direction for projects that span algorithm design, evaluation strategy, and practical landing.
Background
My path combines academic research with product-facing algorithm work across media, vision, and multimodal intelligence.
Current
TikTok
Researcher and algorithm engineer for large-scale audio-video understanding and multimodal intelligence in global content products.
Industry
YY Live (Baidu Group)
Senior Computer Vision Algorithm Engineer leading important algorithm efforts in livestream understanding, content intelligence, and creator-facing visual AI.
Research
Lenovo Machine Intelligence Center
Staff Researcher working on applied computer vision research, reusable tooling, and intelligent systems.
Education
Hong Kong Baptist University, South China University of Technology, Michigan State University
Ph.D. in Computer Science (HKBU, 2018), B.Eng. in Computer Science and Technology (SCUT, 2013), and visiting-scholar experience at Michigan State University.
How I Like To Work
I value clear problem framing, technically defensible decisions, and concise communication. The best projects usually combine curiosity, rigor, and strong ownership while keeping the bar high on algorithm quality.
Algorithm depth
Push on model design, failure analysis, and careful experimentation when a problem deserves real depth.
Direction and ownership
Shape direction, priorities, and evaluation standards so teams can move coherently on difficult problems.
Research-to-product translation
Bridge research, engineering, and product expectations without losing sight of algorithm quality.
Profiles
For a more complete record of publications and activity, these public profiles are the best entry points.