Woong Shin, Ph.D. is a Research Scientist at Oak Ridge National Laboratory, working on agentic AI systems for scientific discovery and intelligent infrastructure for high-performance computing. His current focus includes AI-based systems and workflows for research automation and AI/ML-driven methods for operating exascale supercomputers like Frontier. He holds a Ph.D. in Electrical Engineering and Computer Science from Seoul National University. His work has been recognized with Best Paper at SC21 for HPC energy efficiency research, an R&D 100 Award for the ExaDigiT digital twin framework, and Distinguished Paper at the WORKS workshop (SC25) for research on scientific workflows in the agentic AI era.
Presentation Title:
Fantastic Scientific Agents and How to Build Them: From Workflow (R)evolution to Autonomous Labs
Presentation Abstract:
Autonomous laboratories are the frontier of scientific workflows, but the agents that would run them have neither a construction method nor a coordination substrate. This talk covers both. The revolution the agentic AI era demands of scientific workflows is better run as an evolution, where workflows and agents share a state machine abstraction and progress along two dimensions, intelligence from static to intelligent and composition from single to swarm. Construction comes next. AgentBuild makes agent building a workflow stage, compiling an agent from a scientist-authored contract of rubric, curriculum, and knowledge base, so the build compiles the agent and never the scientist’s judgment. REACH places the intelligence in the orchestration layer itself, with durable execution and idempotency. Based on the case study of Rietveld refinement of X-ray diffraction through GSAS-II, we close on the open problems, drawn from the ORNL Autonomous Chemistry Lab.