The paper describes a multi-agent, retrieval-augmented conversational AI system that provides personalized clinical decision support for Parkinson disease and reported high accuracy on 50 representative clinical queries.

While it offers little in the way of new mechanistic insights, biomarkers, or therapeutic leads, the explainable, scalable tool could improve clinical workflows, patient stratification, and data capture, indirectly aiding translational research and trial recruitment.

Parkinson disease (PD) is a progressive neurodegenerative disorder that poses significant challenges in diagnosis, treatment planning, and long-term care, as patients and healthcare providers often lack timely and context-specific information. This study presents a collaborative multi-agent conversational artificial intelligence system designed to support clinical decision-making and personalized management of Parkinson disease. The system employs generation, critique, and synthesis agents, where generation agents utilize Qwen3-Medical-GRPO, a 4B-parameter medical language model, to produce clinically grounded responses. Critique agents assess factual correctness and clinical relevance, while a synthesis agent ensures coherence and logical consistency. A retrieval-augmented generation (RAG) framework dynamically accesses 80 curated medical resources through a vector-based search engine, integrating user profiling and knowledge graphs to deliver personalized responses. Evaluated on 50 representative clinical queries, the system achieved 95% clinical accuracy, with diagnostic suggestions scoring 4.8/5 and treatment recommendations scoring 4.6/5, and an average response time of 6.5 s. The proposed system provides explainable, scalable, and personalized conversational support, addressing existing gaps in continuity of care, personalization, and accessibility, with the potential to enhance clinical workflows and patient outcomes.