I work on the systems underneath modern AI: high-throughput inference, RL post-training, GPU kernels, and reliable training infrastructure.
Currently, I'm an ML infrastructure engineer at Mind Lab, turning ambitious research ideas into systems that can actually run, scale, and be debugged.
profile.status = "shipping"
profile.interests = ["LLM serving", "RL infrastructure", "GPU kernels", "developer tools"]
profile.tools = ["Python", "CUDA", "Triton", "PyTorch", "Rust", "C++", "Linux"]
| Area | What it means in practice |
|---|---|
| LLM systems | Serving, scheduling, memory, and performance work around vLLM and SGLang |
| RL infrastructure | Training and post-training systems with hardware-aware execution |
| GPU kernels | Triton/CUDA kernels where every microsecond matters |
| Engineering tools | Small, sharp tools that make research workflows easier to operate |
Measure twice. Profile once. Ship often.