Mental health issues in India pose a significant challenge, impacting a large portion of the population due to the country’s vast demographic. According to World Health Organization (WHO), the burden of mental health problems in India amounts to 2443 disability-adjusted life years (DALYs) per 10000 population. Addressing these concerns is critical, as untreated mental health conditions can lead to severe consequences. This paper aims to enhance mental healthcare by developing a task-oriented, closed-domain chatbot that accepts multi-modal input—both voice and text—to provide support and information on mental health issues. We present a comparative study of chatbot models using different embeddings and transformer architectures, focusing on a retrieval-based system built on an LSTM-RNN model. Our LSTM-RNN model achieved an accuracy of 89.36%.