Wooseok Gwak

Hi, I am a first-year PhD student in Computer Science at Georgia Tech, advised by Prof. Anand Iyer in the NEXS group. Before joining Georgia Tech, I received my B.S. in Computer Science (with a minor in Mathematics) from KAIST, where I was a research intern at the CASYS lab advised by Prof. Youngjin Kwon.

My research interests lie in computer systems for machine learning, with a focus on building efficient AI infrastructure — LLM serving systems, memory and KV cache management, and the runtime mechanisms that make them fast and dependable.

Previously, I have worked on research projects including:

• Project S3: a serving system that redesigns the KV caching stack to make sparse attention practical for long-context LLM serving. I focused on reducing CPU–GPU swap overhead when the KV cache is offloaded and reloaded.

• Project DynoSpec: a serving system for speculative decoding that accelerates LLM inference by dynamically colocating models.

• Project TETRIS: a serving system that enables dynamic parallel scaling for LLMs. It assigns adaptive scaling configurations using reinforcement learning.

• Project BudAlloc: One Time memory Allocator (OTA) that mitigates use-after-free (UAF) bugs by decoupling virtual address management from the kernel. It leverages eBPF to reduce the semantic gap.

If you’re interested in discussing research, or exploring collaboration opportunities, I’d love to connect — don’t hesitate to reach out to me at wgwak3@gatech.edu !