This research introduces an AI-driven framework designed for multimodal emotion recognition and sentiment analysis, which combines facial analysis with text-based affective modeling to enhance personalized emotional healthcare. Facial data is analyzed using Deepface to estimate emotions, age, gender, and number of faces, alongside preprocessing methods like face detection, normalization, and alignment. For text analysis, transformer-based models are utilized, specifically a DistilRoBERTa model for recognizing multiple emotions and a RoBERTa model for detecting sentiment polarity. The system includes fallback mechanisms to generate outputs in limited environments by using randomized distributions of age, gender, number of faces, and text-based emotions. The framework was trained and validated using datasets such as FER-2013 and AffectNet, allowing for the identification of various emotions beyond simple binary sentiment. A user interface offers emotion diaries, visual analytics, and long-term mood tracking, providing actionable insights and personalized recommendations. By integrating Deepface-based facial analysis with transformer-based text modeling and incorporating robust fallback strategies, the system moves towards a comprehensive, context-aware, and empathetic AI platform for mental wellness.