何か成す者とは歩み続ける愚者である、成せぬ者とは歩めを止めた賢者である
成一事者，矢志不渝之愚者；毁一事者，停滞不前之贤者
The achievers are the fools who keep walking; the non-achievers are the wise who have stopped.———《ロクでなし魔術講師と禁忌教典》
I'm a Research Scientist at Mind Lab. My work sits at the intersection of LLM training and inference infrastructure — building distributed RL training systems, LLM training/inference engine, optimizing CUDA kernels, and developing self-envolving agents.
I build things across the stack: distributed training systems for RL, CUDA kernels for the hot path, Rust/Go services for the infrastructure layer, and TypeScript tooling for the agent ecosystem.
- Currently building: distributed RL training service, LoRA fused kernel optimization, LLM KV cache storage (Rust), persistent PTY agent tooling
- Core interests: RL training infrastructure, CUDA kernel optimization, distributed systems, LLM inference
- Languages: Rust, Go, TypeScript, Python, CUDA