“Mixture of Doctors”: an agentic-RAG system for self-management of chronic diseases

Supporting patients in the self-management of chronic diseases requires continuous, personalized guidance, a task for which LLM-powered chatbots show immense promise. However, their deployment is contingent on overcoming a dual challenge: the inherent risk of LLM “hallucinations”, which undermines trust, and the architectural complexity of building a system that can manage multiple medical domains securely and reliably. This report introduces “Mixture of Doctors”, an agentic-RAG system designed to address these challenges in tandem. The former is mitigated through a RetrievalAugmented Generation (RAG) flow, where specialized “Doctor” components, each an expert on a specific chronic disease, ground their responses in curated knowledge bases to ensure factual accuracy. The latter is addressed by a robust, decoupled Orchestrator-worker architecture that leverages asynchronous messaging for resilience and open-source technologies for data privacy. The system’s core is an agentic Orchestrator that dynamically analyzes user queries, decomposes complex, cross-domain questions, and routes them to the appropriate specialists. The resulting framework demonstrates a viable path toward creating trustworthy and scalable conversational AI for personalized chronic care.