Closing the Loop: Using MLflow and GEPA to Improve General Supervisor Agents.
Today, organizations are moving beyond simple, single-turn chatbots to multi-agent systems as ways to handle complex, multi-step tasks. Central to these complex systems is the Supervisor Agent, an orchestrator that routes queries, manages sub-agents, and synthesizes final answers. While these new multi-agent system architectures offer their merits, how do you systematically measure, improve, and optimize the system's “manager” when the system is live?
