A Hybrid AI Chatbot Framework for Scalable Mental Health Assistance

Author : Parth Vyas, Aitrayee Bhattacharya

Abstract : The growing global burden of mental health disorders has exposed significant limitations in existing healthcare systems, particularly in providing accessible and scalable support. Artificial intelligence based conversational agents have emerged as a promising solution; however, current approaches face a fundamental trade-off between safety and empathy. Rule-based systems ensure reliability but lack conversational flexibility, while generative models enable humanlike interactions but introduce risks in high-stakes situations. This paper presents a hybrid AI chatbot framework that integrates a BERT-based risk classification module, a rule-based safety system, and a fine-tuned large language model for empathetic response generation. The system dynamically routes user inputs based on real-time risk assessment, ensuring appropriate handling of both crisis and non-crisis interactions. Experimental evaluation on 500 simulated conversations shows that the proposed approach reduces unsafe outputs by 94.3% compared to standalone generative models, while maintaining high levels of user satisfaction and engagement. These findings demonstrate the effectiveness of hybrid architectures in delivering safe, scalable, and context-aware mental health support

Keywords : Artificial intelligence, mental health chatbot, conversational AI, hybrid AI architecture, BERT, large language model, risk classification, rule-based system, empathetic response generation, crisis intervention.

Conference Name : International Conference on Psychology, Psychiatry and Mental Health Integration (ICPPMI-26)

Conference Place : Pune, India

Conference Date : 6th Jun 2026