Orchestrator-Centered Multi-Agent Systems for Clinical Reasoning

Orchestrator-Centered Multi-Agent Systems for Clinical Reasoning

This project develops a multi-agent clinical reasoning system in which specialized large language model agents generate, review, and refine diagnostic predictions through structured multi-step interactions.

The central orchestrator learns how to select agents, assign roles, coordinate discussion, and determine when to stop. We are investigating reinforcement learning methods that balance diagnostic performance, evidence quality, and inference cost, with the goal of building reliable and efficient AI systems aligned with real clinical decision-support workflows.