Mental Health Support Using Gen-AI Shot Prompting Technique and Vector Embeddings

Mental illness has become an issue of concern all over the world, with millions experiencing anxiety, depression, and stress. Chatbots powered by artificial intelligence have proved to be a scalable solution to delivering initial assistance through accessible and instant solutions. This work introduces a chatbot for mental wellness that utilizes generative AI and embedding-based relevance detection to provide contextually relevant and empathic answers. In contrast to rule-based systems, the chatbot utilizes few-shot learning, which enables the system to dynamically adjust according to user queries with a low level of training data. The system uses Sentence Transformer embeddings to determine query relevance and the Google Gemini API to create human-like responses. A series of experiments involving classification accuracy tests, response quality evaluation, and user interaction studies proved the high efficacy of the chatbot in identifying relevant queries while sustaining an empathetic tone of conversation. However, issues like spurious positives to vague inputs and crisis response bottlenecks are still areas that need to be addressed. Areas of future improvement include dataset increase, reinforcement learning, multilinguality, and improving crisis intervention mechanisms to improve the effectiveness of the chatbot in offering mental health support. The chatbot was tested using a curated dataset of mental health dialogues in English, with future work aimed at multilingual expansion.