MS Computer Engineering at NYU, BS Data Science at Duke. I build practical AI systems with a focus on LLM infrastructure, inference performance, observability, and customer-facing AI workflows.

I like work that turns messy model-serving behavior into something measurable: benchmarks, traces, evals, small correctness tests, better docs, and tools that make AI systems easier to operate.

• LLM serving infrastructure: SGLang, vLLM-style serving, llm-d, routing, KV-cache behavior, and OpenAI-compatible APIs.
• Performance tooling: inference benchmarks, kernel correctness checks, latency reports, and lightweight observability.
• Forward-deployed AI systems: support/onboarding agents, RAG workflows, citations, escalation paths, and eval loops.
• Applied AI products: local-first assistants and domain workflows where reliability matters more than demo polish.

• SGLang PR #29205: parser fix and unit test for flat max_dynamic_patch image metadata in Jinja template content.
• llm-d PR #1942: contributor guide for coding agents working inside an AI infrastructure repository.

• Start with a small reproducible path before optimizing.
• Prefer tests, eval cases, and traceable outputs over broad claims.
• Keep setup steps explicit because developer experience is part of the product.
• Write docs and runbooks when they make a system easier for the next person to operate.

• Computer engineering, data science, medical imaging, visualization, and cloud data pipelines.
• Comfortable with Python, JavaScript/TypeScript, Rust, C++/CUDA, SQL, Docker, PyTorch, FastAPI, Three.js, and ML/data tooling.
• Bilingual: English and Chinese.

• LinkedIn
• Email