Development of a Yoruba-Centric Chatbot Utilizing Retrieval-Augmented Generation for Cognitive Behavioural Therapy

Mental health disorders continue to pose significant challenges globally, with limited access to culturally relevant and affordable therapy in many regions, including Nigeria. This project presents the development of a Yoruba-centric chatbot designed to provide cognitive behavioural therapy (CBT) using Retrieval-Augmented Generation (RAG). The solution aims to bridge the linguistic and accessibility gaps in mental health care for Yoruba-speaking populations by integrating large language models (LLMs) with a vector database containing translated and annotated CBT resources in Yoruba. The system leverages Meta AI's No Language Left Behind (NLLB) for data translation, Pinecone for vector storage, and Google's Gemini model for generative responses. Culturally aligned therapeutic content, including Yoruba proverbs and localized dialogue, was used to enhance contextual relevance. The chatbot was deployed via Streamlit, offering an accessible web-based interface for users to receive support in their native language. Evaluation metrics, including ROUGE and METEOR, were applied to assess the accuracy, coherence, and semantic relevance of responses. The results demonstrate that the chatbot effectively delivers contextually appropriate mental health support while preserving cultural and linguistic authenticity. This study highlights the potential of Artificial Intelligence (AI) in democratizing access to mental health services for underrepresented language communities and sets a framework for expanding such interventions to other African languages and cultures.