This workshop will be held in the Data Hub (Library room 107) and live via Zoom.

In this talk, we’ll dissect agentic system architectures from tool-calling interfaces and the Model Context Protocol (MCP) to direct computer-control integrations and multi-agent coordination patterns. We'll explore how existing agentic frameworks handle orchestration, communication protocols, human-in-the-loop safety, evaluation metrics, failure modes, and security considerations. We’ll examine agentic models like Kimi K2, specifically trained for optimized task orchestration, and how they slot into broader pipelines. Finally, we’ll cover practical approaches to monitoring, logging, resource management, and scaling multi-agent deployments in production.