M.S. student at Yonsei University ASO Lab, working on AI systems — deep learning compilers, GPU kernel optimization, and LLM inference serving.
- Deep learning compilers — tensor program auto-tuning, TVM, Triton,
torch.compile - LLM inference serving — vLLM / SGLang internals, speculative decoding, KV-cache management
- LLM-guided systems optimization — using LLMs to steer compiler and kernel search
| Project | What it does |
|---|---|
| adaptive-transfer-reasoning-compiler | Transfer memory for low-budget matmul auto-tuning — 1.188× compiled-runtime speedup |
| profbridge-llm-guided-gpu-kernel-optimization | Profiler-aware evaluation of LLM-generated GPU kernel candidates |
| shared-mcts-compiler-optimization | Multi-model shared MCTS for TVM tensor program optimization |
| rgta-mtmapd | Regret-guided online task-group assignment for multi-agent pickup & delivery |
- Blog (Korean): yibangwon.tistory.com
- Contact: hansol.son@yonsei.ac.kr