MindMate: An Emotion-Aware Generative AI System for Personalized Mental Health Support

Abstract

We present MindMate, an AI-powered mental health assistant that combines a fine-tuned DeepSeek-R1 language model with BERT-based emotion recognition to deliver personalized therapeutic dialogues. The system analyzes user inputs in real-time (84% emotion detection accuracy) and generates contextually appropriate responses while identifying crisis situations (91% recall). Implemented on Google Colab Pro+ using 4-bit quantization, MindMate achieves 83% user satisfaction in trials with 30 participants, demonstrating comparable performance to commercial mental health Chatbots. The architecture's novel integration of generative AI with clinical knowledge bases enables accessible, emotionally intelligent support while maintaining response quality. This work provides a blueprint for developing effective, open-weight mental health assistants without proprietary dependencies.