Cognicare : An AI-Powered Conversational Agent for Mental Health Monitoring and Support

Mental health disorders, including depression, anx- iety, and suicidal ideation, present significant challenges for continuous care, as symptoms often evolve undetected between clinical visits. This paper introduces Cognicare, an AI-driven conversational agent designed for real-time emotional monitoring, longitudinal risk assessment, and clinically actionable insights. Cognicare combines a fine-tuned RoBERTa model for multi-class mental health classification with DistilBERT-based sentiment analysis. These outputs are fused via a Dynamic Distress Scor- ing Algorithm, generating personalized, context-aware distress metrics that account for linguistic cues, temporal trends, and model confidence.Therapeutic interactions leverage a large lan- guage model (LLM) aligned with Cognitive Behavioral Therapy principles through structured prompt-chaining, ensuring emo- tionally congruent, contextually relevant, and psychologically safe responses. The system tracks longitudinal emotion trajectories, detects anomalies, and produces HL7 FHIR-compliant reports for clinicians, highlighting high-risk cases and trend patterns to support timely interventions.Evaluations demonstrate improved classification accuracy, enhanced empathy, and reduced toxicity in generated responses. Cognicare illustrates how integrating advanced NLP models with clinically informed design can pro- vide scalable, accessible, and reliable continuous mental health support, bridging the gap between user self-expression and evidence-based care.