Title: LLM Agent Systems: From Structured Workflows to Long-Horizon Discovery
Date: November 25, 2026
Time: 9:00–11:00 AM Eastern Time
Location: Coda C1015 Vinings;  https://teams.microsoft.com/meet/2951218374790206?p=DDMohIcZYEev3VqrHv
Bo Yuan
Machine Learning PhD Candidate
Daniel Guggenheim School of Aerospace Engineering
Georgia Institute of Technology
Committee
1.    Dr. Yongxin Chen (Advisor), Georgia Institute of Technology
2.    Dr. Kartik Goyal, Georgia Institute of Technology
3.    Dr. Wei Xu, Georgia Institute of Technology
4.    Dr. Matthew Gombolay, Georgia Institute of Technology
5.    Dr. Bin Hu, University of Illinois Urbana-Champaign
Abstract
Large language models (LLMs) open new possibilities for automating complex tasks in engineering and scientific research, yet realizing them takes more than answering individual questions. This dissertation studies the resulting challenges in four settings, namely workflow construction, multi-agent collaboration, interactive design, and long-horizon discovery. Finally, a case study turns to LLM agents on open mathematical problems, focusing on questions in high-dimensional sampling. It shows how LLM agents, working alongside human researchers who provide guidance and corrections, can help resolve research-level open problems.