Agentic Simulation for Emergency Department Operations

EDSim models an emergency department as a population of autonomous agents. Patients arrive with complaints and histories, nurses triage and monitor them, and physicians order tests and decide on disposition, each agent driven by a large language model that reads the state of the department and acts on it. The simulation covers the whole floor rather than a single encounter, so the queues, handoffs, and waiting that shape a real shift emerge from what the agents do to each other.

Simulating the department makes it possible to try a change before it reaches patients. Staffing mixes, triage rules, and bed allocation can be varied across runs to show where the bottlenecks sit and which changes move the numbers, work that is hard to do in a live department without disrupting care.

EDSim sits alongside the lab’s other work on emergency care, including a benchmark of small language models for emergency department decision support and a study of how human-aware agentic artificial intelligence could reshape emergency care workflows. The simulator is open source, with a short video walking through a full run.