A two-agent structure for enhanced reasoning
Our work addresses this problem with a novel strategy primarily based on the interaction of two LLM-driven brokers, which is similar to how human clinicians deal with administration issues.
The Dialogue Agent is user-facing and geared up to quickly reply primarily based on its present understanding of the affected person. This agent handles the conversational points of the interplay, gathering details about the affected person’s situation, addressing their considerations, and constructing rapport. By leveraging pure language processing and empathetic communication strategies, the Dialogue Agent ensures a seamless and interesting person expertise.
The Mx Agent (Administration Reasoning Agent) intentionally and constantly analyzes the accessible info, together with medical tips and patient-specific information, to optimize administration of the affected person. Leveraging Gemini’s state-of-the-art long-context capabilities, this agent synthesizes and causes over giant quantities of knowledge — affected person dialogues throughout a number of visits along with a whole bunch of pages of medical tips — unexpectedly. Utilizing this strategy, it produces structured plans for investigations, therapies, and follow-up care, considering the newest medical proof, info gathered throughout earlier visits, and particular person affected person preferences.