HEALX: Agentic Health Monitoring with IoT, RAG, and Cloud-Native Multi-Agent AI

Abstract

HEALX is a cloud-based health monitoring system that uses wearable sensors, artificial intelligence, and medical knowledge retrieval to provide users with preliminary health-risk information. The system is based on a wearable device on ESP32 to collect vital physiological parameters such as heart rate, blood oxygen saturation (SpO2), and body temperature. These readings are sent to a backend based on FastAPI where they are processed along with the symptoms provided by the user. HEALX uses four specialised AI agents that provide a more structured process for health assessment: Symptom Interpreter, Symptom Summariser, Clarification Agent and Risk Assessment Agent. These agents work together, not unlike a single AI prompt. Language-based processing is done using Llama-3.3-70b-versatile via the Groq API. AI agents generate responses with relevant medical information provided by a Retrieval-Augmented Generation (RAG) mechanism using ChromaDB and the text-embedding-3-small model. One of the key features of HEALX is that the AI reasoning is not allowed to act independently of the sensor readings. Critical readings can bypass the normal AI reasoning process by applying programmatic safety rules to physiological values. The whole system is built with AWS services, such as a Virtual Private Cloud (VPC), Application Load Balancer, EC2 Auto Scaling, Amazon SQS, Redis, and Supabase PostgreSQL. React/Vite based web interface. Supports English, Hindi and Marathi. HEALX achieved an average end-to-end response time of 8.2 seconds, vector retrieval latency of <50 milliseconds, reported cloud uptime of 99.7%, and 90% risk-stratification accuracy in 20 validated cases in the reported simulations. Human annotators evaluated the follow-up questions, assigning them an average relevance score of 4.3 out of 5. The system is designed to help with preventive screening and health-related decision support and is not meant to replace professional medical diagnosis.