2026 KDD Hands-on Tutorial
Tuesday, August 11, 2026 1:30 PM – 4:30 PM
Location: Jeju, Korea
Industrial asset operations and maintenance require decisions grounded in heterogeneous data, including sensor streams, asset hierarchies, failure-mode catalogs, and work-order histories. Large language model agents can unify these sources and automate workflows such as condition monitoring, fault diagnosis, and maintenance planning, but deployment in safety-critical environments demands reliable tool use, traceable reasoning, and rigorous evaluation. This tutorial introduces the Model Context Protocol (MCP) as a practical foundation for connecting agents to industrial systems and AssetOpsBench as a reproducible environment for developing and benchmarking them.
Participants will progress from core MCP concepts to hands-on execution with tools for utilities, IoT and asset registries, failure-mode retrieval, work orders, time-series analysis, and vibration diagnostics. Using guided notebooks, they will build a grounded work-order agent with Stirrup, inspect its tool-use trajectory, evaluate its outputs against static scenario data, and compare systems on an auditable leaderboard. The tutorial connects architectural choices with grounding, safety, provenance, robustness, and operational cost, equipping researchers and practitioners to design, diagnose, and evaluate reliable AI agents for real-world industrial operations.