AI Papers Reader

Personalized digests of latest AI research

View on GitHub

On-Device AI Medical Assistant Revolutionizes Healthcare with Multi-Agent Architecture

A team of researchers from Samsung Research Institute Bangalore has developed a groundbreaking on-device, multi-agent healthcare assistant designed to address the limitations of existing cloud-based solutions in healthcare. Their work, detailed in a recent preprint, introduces a system that prioritizes user privacy, reduces latency, and ensures reliable operation even without internet access. This innovative approach combines the power of large action models (LAMs) with a distributed, multi-agent system optimized for edge devices like smartphones and smartwatches.

The core challenge addressed by the research is the inherent conflict between the advantages of large language models (LLMs) in healthcare and the practical constraints of deploying them on resource-limited edge devices. LLMs, while excellent at complex tasks, often demand substantial computational power and continuous internet connectivity, raising privacy concerns regarding sensitive patient data.

The Samsung team’s solution cleverly bypasses this problem using a multi-agent system. Instead of relying on a single, massive LLM, they employ several smaller, specialized agents, each responsible for a specific task within the healthcare workflow. This approach allows for optimized resource allocation and improved scalability—adding new functionalities simply requires integrating new agents. The system’s architecture comprises three key components:

  • Action Manager: This orchestrates the interaction between the agents, managing task planning and execution.
  • Health Manager: This houses several agents that independently handle tasks like health monitoring, report generation, and appointment scheduling. For example, one agent might monitor vital signs from a smartwatch, while another creates daily health summaries.
  • Memory Unit: This maintains both short-term (for the current interaction) and long-term (for patient history and preferences) memory, allowing for personalized experiences and informed decision-making by the agents.

The system incorporates several key functionalities, including:

  • Intelligent Diagnosis and Appointment Scheduling: The system engages in interactive conversations with the user to gather comprehensive health information, intelligently identifying the most appropriate medical specialist based on the user’s symptoms and history.
  • Emergency SOS: Users can manually trigger an SOS signal in emergencies. The system then automatically contacts emergency services and designated contacts, providing real-time location information.
  • Vitals Tracking and Abnormality Detection: The system monitors vital signs from wearables and triggers alerts for potential problems.
  • Medication Reminders and Daily Health Reporting: The system automates medication reminders and generates daily health summaries.

The researchers trained their agents using a combination of fine-tuning and prompt engineering techniques. Notably, they developed a sophisticated data creation pipeline to generate synthetic datasets for training. The use of synthetic data helps to address the challenges associated with acquiring and managing sensitive patient data for training purposes. The fine-tuned planner and caller agents achieved high RougeL scores (85.5 for planning and 96.5 for calling), demonstrating effective task completion using their generated data.

This innovative approach represents a significant advancement in on-device AI for healthcare, offering the potential for personalized, privacy-preserving, and always-available healthcare assistance. The research team plans to further enhance the system by deploying it on mobile devices and expanding its functionalities to include more sophisticated health monitoring and diagnostics. This on-device solution promises to transform how individuals manage their health and access crucial medical information, paving the way for a more user-centric, efficient, and secure healthcare system.