Multimodal Agentic Assistants for Primary Care

This project centered on the end-to-end design, scientific leadership, and production deployment of a patient-facing, agentic AI assistant for primary care at Amazon One Medical. The system supports natural language interactions across text and voice, understands a member’s longitudinal health context, and can safely take action (such as guiding care decisions, routing to clinicians, and initiating workflows) within a regulated healthcare environment.

Unlike traditional healthcare chatbots, this assistant was built as an agentic system: it combines large language models with structured decision logic, clinical safety layers, and healthcare-specific tooling to move beyond Q&A toward trusted, outcome-oriented assistance. The work required balancing cutting-edge multimodal AI with the realities of medical risk, compliance, and real-world deployment at scale.

The project launched publicly as Amazon One Medical’s Health AI assistant, marking one of the first large-scale deployments of an agentic, healthcare-native AI system embedded directly into a primary care experience.

Media coverage

Pictures