Title: Multi-Agent Systems: Design Principles and a Case Study in AI Pipeline Generation

Speaker: Dr. Yunsu Kim @ aiXplain

Time : 16:30 ~ 17:30, April 10th, 2025

Location: Online

• In-person: X

• Online: https://hli.skku.edu/InvitedTalk250410  

  • Meeting ID: 863 6694 9104

  • Passcode: 750391

Language: English speech & English slides

Abstract: 

Multi-agent systems enable autonomous agents to collaborate, adapt, and optimize workflows in complex environments. This talk explores their core components, workflows, and design paradigms, including hierarchical, distributed, reactive, and self-learning approaches. We then present a case study on Bel Esprit, a multi-agent system for AI pipeline generation, detailing its task definition, system architecture, key components (Mentalist, Builder, Inspector, Matchmaker), and experimental results. The audience will gain a practical understanding of multi-agent system design and its real-world applications.

Bio:

Yunsu Kim holds a Ph.D. from RWTH Aachen, Germany, specializing in neural machine translation for low-resource scenarios under the supervision of Prof. Hermann Ney. During his Ph.D. studies, he also worked as a Machine Translation Scientist at AppTek, building enterprise translation models, and served as a research coordinator between eBay and RWTH Aachen. He achieved first place at WMT 2018 in the supervised, unsupervised, and corpus filtering tracks. He later became a Senior Research Scientist at Lilt, focusing on post-editing and optimizing translation suggestions. As an Assistant Professor at POSTECH, he researched translation post-editing, question answering, essay scoring, speech recognition and synthesis, dialog state tracking, and text summarization. Currently, he is a Senior Applied Scientist at aiXplain, focusing on developing chatbot agents for model pipeline creation.