CT Radiology Triage

Our project is a multi-agent orchestration platform designed to optimize the high-stakes environment of a radiology CT queue. Recognizing the critical need for speed and accuracy in regulated healthcare settings, we have built a system where five specialized agents collaborate through the Band framework to optimise this workflow. The workflow begins with the ct_dispatcher_agent which creates a dedicated Band room for every incoming clinical case, ensuring isolated and structured communication. The ct_router_agent then normalizes the data, preserving vital patient metadata while performing initial urgency assessments to prepare the case for review. The core intelligence of the system lies in the interplay between the ct_review_agent and ct_moderator_agent. These LangGraph-powered agents utilize LLMs to perform deep clinical reviews and strategic queue comparisons. Specifically, the Moderator agent determines the optimal queue position through a binary-search pairwise comparison against the current queue context, ensuring the most urgent patients are prioritized based on clinical reasoning. For cases with high uncertainty, invalid payloads, or clinical red flags, the ct_escalation_agent triggers a structured human-in-the-loop handoff, providing clinicians with a comprehensive audit trail and explicit choices for intervention. By using Band as the underlying coordination layer, we seamlessly integrate disparate frameworks—combining standard Python logic with complex LangGraph workflows—to deliver a traceable, efficient, and enterprise-ready solution that reduces manual coordination and accelerates life-critical medical decisions.