Research

My current research profile is organized on the home page, with selected publications listed under Publications.

Current Focus

Cross-Layer Characterization of GPU LLM Workloads

I study how GPU compute kernels, communication libraries, runtime scheduling, and hardware resources interact in distributed LLM workloads. My recent work shows that compute-communication overlap can introduce hidden hardware-level costs even when it improves apparent pipeline utilization.

Hardware-Software Co-Design for Efficient AI Systems

I am interested in systems that are designed across model structure, runtime behavior, communication, and architecture rather than optimized one layer at a time.

Contact: jihwanoh@stanford.edu