AutismCarebot: An Empathy-Enhanced and Retrieval-Augmented Chatbot for Autistic Users

Abstract

Many autistic individuals require specialized communication strategies, tailoredemotional understanding, and personalized resources - areas often overlooked by general mental health chatbots. This thesis presents AutismCarebot, an empathy-enhanced, retrieval-augmented conversational AI built upon the LLaMA-3.2 (3B) Instruct model, designed specifically to support autistic users. The model was initially fine-tuned on empathetic dialogue datasets (Empathetic Dialogues, Amod Mental Health Counseling Conversations, Psych8k, ExTES) and further adapted using autism-specific data (TASD) to better align with autistic communication styles. The chatbot uses keyword-based heuristics - simple rule-driven methods - to detect emotional cues such as anxiety, sadness, and overwhelm, responding with clear, empathetic validation and tailored reassurance. Retrieval-Augmented Generation (RAG) helps improve factual grounding by embedding evidence-based advice from credible autism-related resources, with transparent citations. A keyword-triggered crisis support feature connects users to global helplines when self-harm indicators are detected. Usability testing and qualitative feedback from autistic individuals, caregivers, and educators indicated positive perceptions of empathy, emotional responsiveness, and trust. AutismCarebot demonstrates a promising, ethically aligned approach to enhancing emotional support and crisis safety for autistic users, with potential applications in therapeutic, educational, and peer-support contexts.