hemhemoh/Wazobia-Wellness

Wazobia-Wellness is a multilingual mental health chatbot that provides culturally-sensitive therapeutic support in English, Yoruba, Igbo, and Hausa using fine-tuned Gemma-2-2b-it and NLLB-200 translation models. The project addresses language barriers in mental healthcare access for Nigerian communities. # Wazobia-Wellness: A Multilingual Mental Health Chatbot for Nigerians ## 🌍 Project Overview **Wazobia-Wellness** is a multilingual mental health chatbot specifically designed to serve Nigerian communities in their native languages. The name combines "Wazobia" - a term that means "come" in three major Nigerian languages: Yoruba (wa), Hausa (zo), and Igbo (bia), often used as a symbol of unity, diversity, and inclusion in Nigeria - with "Wellness" to emphasize the project's mission of making mental health support accessible across Nigeria's diverse linguistic landscape. This project addresses the critical need for culturally and linguistically appropriate mental health resources in Nigeria, where over 250 ethnic groups and languages exist, but mental health services are predominantly available only in English. ## 🎯 Project Objectives - **Linguistic Inclusivity**: Provide mental health support in English, Yoruba, Igbo, and Hausa to ensure users can communicate in their most comfortable language - **Cultural Sensitivity**: Deliver culturally appropriate mental health guidance that respects Nigerian values and contexts - **Accessibility**: Break down language barriers that prevent many Nigerians from accessing mental health support - **AI-Powered Support**: Leverage advanced language models to provide empathetic, knowledgeable responses to mental health concerns ## 🛠️ Technical Architecture ### Base Model - **Foundation**: Google's Gemma-2-2b-it (2 billion parameter instruction-tuned model) - **Fine-tuning Approach**: Parameter-Efficient Fine-Tuning (PEFT) using LoRA (Low-Rank Adaptation) - **Quantization**: 4-bit quantization with BitsAndBytesConfig for memory efficiency ### Multilingual Translation Pipeline - **Translation Model**: Facebook's NLLB-200-distilled-600M (No Language Left Behind) - **Target Languages**: Hausa (hau_Latn), Igbo (ibo_Latn), Yoruba (yor_Latn) - **Processing**: Batch translation with intelligent text chunking for long conversations ### …