Unique Presentation Identifier:

14

Faculty Advisor

Dr. Robin Ghosh

Document Type

Poster

Location

Face-to-face

Start Date

9-4-2026 1:00 PM

End Date

9-4-2026 3:00 PM

Abstract

Mental health challenges such as anxiety and depression are widespread, yet often remain unaddressed due to stigma, financial barriers, limited access to professionals, or personal reluctance. To address this gap, we present an accessible AI-powered chatbot that provides preliminary conversational support with empathetic, context-aware responses. The chatbot is trained and evaluated on two publicly available counseling conversation datasets from Huggingface, enabling it to learn and maintain therapeutic dialogue patterns. We compare three advanced LLMs, such as Llama3.1, Mistral 7b, and Qwen3, evaluating them on relevance, empathy, conciseness, and contextual understanding, and all models demonstrate high response quality. Incorporation of a Retrieval-Augmented Generation (RAG) pipeline with Facebook AI Similarity Search (FAISS) enhances factual accuracy and contextual relevance while reducing hallucinations. A key innovation is the integration of agentic AI, empowering the chatbot to recognize extreme distress and, with user consent, assist in scheduling appointments with mental health professionals. Overall, our system demonstrates the potential to deliver reliable, empathetic, and actionable conversational AI support for individuals experiencing early-stage mental health distress.

Share

COinS
 
Apr 9th, 1:00 PM Apr 9th, 3:00 PM

Proactive Mental Health Assistance via Agentic LLM Chatbots with Retrieval-Augmented Generation

Face-to-face

Mental health challenges such as anxiety and depression are widespread, yet often remain unaddressed due to stigma, financial barriers, limited access to professionals, or personal reluctance. To address this gap, we present an accessible AI-powered chatbot that provides preliminary conversational support with empathetic, context-aware responses. The chatbot is trained and evaluated on two publicly available counseling conversation datasets from Huggingface, enabling it to learn and maintain therapeutic dialogue patterns. We compare three advanced LLMs, such as Llama3.1, Mistral 7b, and Qwen3, evaluating them on relevance, empathy, conciseness, and contextual understanding, and all models demonstrate high response quality. Incorporation of a Retrieval-Augmented Generation (RAG) pipeline with Facebook AI Similarity Search (FAISS) enhances factual accuracy and contextual relevance while reducing hallucinations. A key innovation is the integration of agentic AI, empowering the chatbot to recognize extreme distress and, with user consent, assist in scheduling appointments with mental health professionals. Overall, our system demonstrates the potential to deliver reliable, empathetic, and actionable conversational AI support for individuals experiencing early-stage mental health distress.