William Lee

Led development and performance optimization of AI inference infrastructure supporting large-scale LLM workloads across GPU-accelerated systems. Built and improved inference runtime components in C++, Rust, and CUDA with emphasis on request scheduling, memory efficiency, reliability, and production performance. Improved model-serving efficiency through batching, KV-cache management, quantization, CUDA graphs, and workload-aware scheduling. Partnered across model, framework, runtime, networking, and GPU teams to identify bottlenecks and turn system-level findings into production improvements.

Designed and developed high-performance C++ and CUDA software for GPU-accelerated compute and machine learning workloads. Investigated performance issues across runtime, memory, compute, and communication layers to improve efficiency and stability of GPU-based systems. Collaborated with hardware, platform, and software teams to deliver production-ready features and support new GPU capabilities across distributed environments.

Led design and delivery of distributed platform components balancing performance, reliability, security, scalability, and operating cost. Developed performance-sensitive systems software in C++ and Rust for cloud infrastructure and distributed workloads. Owned complex engineering work end-to-end from technical design through production rollout, troubleshooting, and ongoing operational support. Guided engineers through design and code reviews and collaborated with networking, security, and product teams on platform improvements.