세미나안내
Towards Heterogeneous LLM Serving in the Era of Agentic AI
- 등록일2026.11.09
- 조회수27
-
세미나 일정2026.11.11 수
-
연사박종세 교수(KAIST)
[Abstract]
Agentic AI is increasing the scale and diversity of large language model (LLM) workloads, creating new challenges in computation, memory, and serving efficiency. This talk explores how heterogeneous hardware and serving software can jointly address these challenges. I will introduce LLMServingSim, a unified simulation infrastructure for evaluating heterogeneous and disaggregated LLM serving systems, and Simthesizer, an agent-driven framework that translates high-level system descriptions into executable simulations. I will then present NELSSA, a GPU–PNM system that uses length-based request placement and dynamic migration to efficiently serve mixed-length workloads. Together, these efforts demonstrate how simulation and hardware–software co-design can guide the development of scalable LLM serving infrastructure, motivating a broader vision for a unified multi-accelerator runtime in the agentic AI era.
[Biography]
Jongse Park is an Associate Professor in the School of Computing at KAIST, where he leads the Computer Architecture and Systems Laboratory (CASYS). His research focuses on building HW-SW co-designed systems for Artificial Intelligence (AI). He is also the Founder and CEO of AnyBridge, which develops runtime system software for efficient LLM and Agent serving across heterogeneous hardware platforms. Prof. Park earned his PhD in Computer Science from Georgia Institute of Technology. Prior to joining KAIST, he worked as an architect at Bigstream, a Silicon Valley-based AI system startup. He has published over 47 refereed articles in top-tier conferences and journals spanning computer architecture, computer systems, database systems, digital circuits, and software engineering, with over 4,500 citations. His work has been recognized with several awards, including Best Paper Award at ISPASS 2026, Distinguished Artifact Award at ISCA 2024, Best Paper and Distinguished Artifact Awards at IISWC 2024, Best Paper Award at IEEE Micro 2024, two ISCA 25-Year Retrospective Awards (1996-2020) in 2023, Distinguished Paper Award at HPCA 2016, and two IEEE Micro Top Picks honorable mentions in 2025 and 2015. Prof. Park has been a program committee member for top-tier conferences such as ISCA, MICRO, ASPLOS, HPCA, SC, DAC, ISPASS, and IISWC.
