Coding Agent and Kernel Agent Infrastructure
Built coding-agent infrastructure that improved IPD efficiency by more than 100%, with a kernel agent achieving first place in the 2026 MLSys Challenge.
Agentic AI · Physical AI · Intelligent Systems
From agentic intelligence to the physical world.
I connect foundation models, agent systems, and efficient infrastructure to move AI from reasoning and planning into perception, action, and real-world deployment.
“A craftsman who wishes to do good work must first sharpen his tools.”
Confucius · The Analects
Profile
Intelligence becomes useful when it can perceive, decide, and act.
I work across the stack that makes this transition possible: foundation-model post-training for reasoning, agent infrastructure for planning and tool use, multimodal systems for perception, efficient inference for real-time action, and hardware-software co-design for physical deployment.
More about meFeatured projects
Systems that carry AI from reasoning into action.
Latest technical notes
Ideas for intelligence that can reason, perceive, and act.
Understanding KV Cache Compression
A systems view of why KV cache becomes the limiting resource in long-context inference and where compression actually helps.
Read noteAgent Memory Is a System Architecture
A practical model for separating working context, durable knowledge, task state, and execution history.
Read note