I work on LLM serving systems. Before that, ~2.5 years building distributed backends at Arista and an internship on LinkedIn's compute capacity platform. MCS @ UIUC (Dec 2026).
Projects
- Mini-SGLang: rewrite of sgl-project/mini-sglang: scheduling, radix KV cache, chunked prefill, overlap scheduling
- Agentic RL for a coding agent: verl GRPO + vLLM rollouts on 8×A6000; LiveCodeBench v6 34.86% → 42.29%
- HeteroMega (in progress): iGPU–NPU tensor parallelism on AMD Ryzen AI Max+ 395
Before: LinkedIn (Compute Broker) · Arista Networks (CloudVision Studio)

