Using Conversational Agents in Smart Care Environments

Abstract

This thesis explores the integration of conversational agents (CAs) and physiological sensing for intelligent, context-aware monitoring in Smart Care Environments. As the global population ages and demands for independent living rise, there is a growing need for systems that can combine natural human–AI interaction with real-time health awareness. The work investigates how multimodal data, particularly movement, electrocardiogram (ECG), galvanic skin response (GSR), and electroencephalogram (EEG) signals, can be incorporated into large language model (LLM)-based dialogue systems to improve emotional sensitivity and personalization. A prototype Smart Care system was developed, consisting of a physiological signal processing pipeline, a stress-detection classifier, and a conversational interface powered by an LLM. An evaluation comparing baseline dialogues and sensor-enhanced prompts demonstrated that including physiological context improved empathy, relevance, and contextual appropriateness in generated responses. The results highlight the potential of combining physiological sensing with conversational AI to enhance adaptive assistance and emotional intelligence in home-based healthcare systems.