Agentic AI System for Stress Monitoring: A Multi-agent Healthcare Crew

Chronic stress poses serious health risks, but traditional detection methods like self-reporting or infrequent assessments lack timeliness and personalisation. Reactive and rigid AI systems further limit adaptability and contextual accuracy, leading to false positives and reduced clinical trust in connected healthcare. Agentic AI represents a paradigm shift by integrating autonomous decision-making, dynamic planning, and adaptive reasoning to operate with minimal human intervention. These systems possess real-time situational awareness, set goals, and refine strategies through self-reflection and memory, making them adaptable across robotics, autonomous vehicles, and decision-support systems. This study explores a multi-agent crew framework with a model agent and a clinical summary agent for real-time monitoring and personalised detection of chronic stress using physiological signals (heart rate and respiratory rate) from the Stress-Predict dataset collected via wearable sensors from 35 participants. The model agent autonomously processes data, extracts features, classifies stress levels using Random Forest and AdaBoost, and applies dynamic thresholding for personalised abnormal stress detection. When stress exceeds the threshold, the clinical summary agent generates a structured clinical report using the Mistral model from Ollama, providing insights and recommendations for stress management. On the binary classification task, Random Forest averaged 97% accuracy compared to 82% for AdaBoost. Dynamic thresholding distinguished acute from chronic stress, generating reports for 61% of participants. This research shows how agentic AI enhances personalised healthcare, improves early stress detection, and addresses transparency and ethics. It serves as a foundation for advancing remote patient monitoring and personalised healthcare interventions by integrating them within a healthcare dataspace.