Stress is a physiological and psychological response to threatening, challenging, or demanding situations. It can have profound effects on both mental and physical health such as anxiety and depression. Thus, managing stress levels in mental healthcare is crucial for maintaining overall well-being. In this work, we introduce SereniSens, a multimodal AI framework with conversational agent integration to predict stress levels during sleep using physiological data. We evaluated the performance of four machine learning models - Decision Tree, Support Vector Machines, Multilayer Perceptron, and XGBoost - on the SaYoPillow dataset containing physiological signals recorded during sleep. Our models achieve high accuracy in predicting fve stress levels, with SVM obtaining 99.37% accuracy, outperforming previous implementations. We also propose an architecture integrating these predictive models with a Large Language Model to create a context-aware chatbot for stress monitoring and management. This framework demonstrates the potential of AI in mental healthcare, particularly for personalized stress assessment and intervention. [ABSTRACT FROM AUTHOR]