MULTI-AGENT AI ASSISTANT FOR RURAL HEALTHCARE SYSTEM

The natural-cause mortality rate (NCM), or the mortality rate attributed to disease, between 1999 and 2019, for 25-54 year-olds in urban areas decreased from 179 per 100,000 citizens to 143, but the NCM for the same demographic in rural areas increased from 189 per 100,000 citizens to 205. Hence, there is a need to address the disparities in rural healthcare. Critical needs of the rural healthcare system and challenges like workforce shortages, limited health literacy, resource scarcity, and adverse socio-economic conditions are highlighted. AI- driven automation presents a viable solution to address these issues. Using a pre- to post- doctor visit framework, these challenges are mapped into specific healthcare tasks that can be automated through AI. A survey of the healthcare AI literature reveals a notable lack of research on rural healthcare tasks. Therefore, we advocate for a dedicated effort to develop AI to address rural healthcare needs. We also argue that educational and administrative tasks, alongside diagnostic tasks, must be prioritized for a holistic approach to improving rural healthcare. Additionally, we argue that phone-call or low-bandwidth based AI voice assistant solutions provide a viable way to overcome rural healthcare service delivery challenges. To address persistent disparities in rural healthcare-such as workforce shortages, low health literacy, and limited digital infrastructure-we present a phone-call-based AI voice assistant tailored for low-bandwidth environments. Built on a multi-agent architecture, specialized LLMs collaborate to handle distinct healthcare tasks and respond to patient queries while supporting providers. The assistant automates administrative workflows like scheduling and documentation, and improves patient literacy by explaining visit details, conditions, and insurance. It is scalable, privacy-preserving through local deployment of open-source models, and includes safety guardrails essential for medical use in rural settings.