AI Infrastructure Engineer building open-source systems for large-model inference, multimodal generation, and reinforcement-learning post-training.
I am a maintainer of vLLM-Omni, VeRL-Omni, and vime. I also lead the vLLM Bilibili channel, where I work with the community to make AI infrastructure easier to understand and contribute to.
- vLLM-Omni — Efficient inference and serving for omni-modality and generative models, including text, image, video, and audio workloads.
- VeRL-Omni — A reinforcement-learning training framework for diffusion and omni-modality models.
- vime — A vLLM-native post-training framework for scaling reinforcement learning.
My work focuses on the systems layer connecting model capability to real products: inference engines, rollout systems, distributed execution, memory management, and high-performance serving.
- 📺 Every Saturday: I host a technical session on the vLLM Bilibili channel, covering inference, post-training, kernels, distributed systems, and emerging AI infrastructure projects.
- 🎓 Every Sunday: I organize an in-person study group at UTown in Singapore, where learners from different backgrounds study, build, and contribute together.
- 📖 Technical notes: I turn talks and slide decks into searchable Chinese and English articles on my AI Infrastructure Notes.
We are living through one of the most consequential technological shifts in human history. AI is already changing software engineering, research, and the way people live and work. For me, the opportunity of this era is not only to build a career, but also to help more people participate in moving it forward.
AI infrastructure is the bridge between model capability and real-world impact. From model deployment and reinforcement-learning rollouts to agents and online serving, systems such as vLLM are becoming essential infrastructure. They are also one of the most accessible entry points into the field: open, practical, and shaped by contributors from many different backgrounds.
I do not believe AI infrastructure should remain in the hands of a small group of specialists. The people who understand the real problems in medicine, education, law, manufacturing, civil engineering, and every other domain should be able to understand and shape the systems that bring AI into their work.
That is why I spend my weekends teaching and organizing study groups: to lower the barrier to AI infrastructure, help people take their first step, and create more opportunities for them to contribute. When more people can understand and build this infrastructure, AI becomes less of a tool controlled by a few and more of a technological transition whose benefits can be shared by society as a whole.