<p>In this paper research work present a smart learning interaction system, which combines multimodal affect recognition with adaptive content selection to enable affect-sensitive and personalized learning. The proposed system recognizes the emotional condition of the learners using three major input modalities, which are facial expression, voice input and text response. Convolutional Neural Networks (CNNs) are used to recognize emotions on the face, and Mel-Frequency Cepstral Coefficients (MFCC) and Bi-directional Long Short-Term Memory (BiLSTM) networks are applied to analyse speech signals. Also, a BERT-based transformer model is used to interpret the textual inputs of the learners to obtain the contextual sentiment. These multimodal emotional cues are combined with a decision level ensemble method in order to enhance precision and trustworthiness. After identifying the emotional state, an adaptive delivery engine powered by reinforcement learning dynamically adapts the learning content, such as the adjustment of the difficulty level, format (e.g., visual, auditory, textual), and pacing to match the emotional and cognitive state of a learner. The system was piloted in a controlled e-learning environment with a sample student group. Evaluation outcomes showed significant improvement in engagement, understanding, and user satisfaction, measured through post-session surveys, quiz performance, and interaction logs compared to conventional passive learning systems. The study implies the possibility of integrating affective computing and intelligent tutoring systems to make educational technologies more humane and effective, which will move the future of human-centred digital learning systems.</p>