Psychological stress significantly affects cognitive performance, emotional stability, and overall well-being among students and professionals. Traditional stress detection methods rely on self-reported surveys or physiological sensors, which are often intrusive, costly, and unsuitable for continuous monitoring. This paper proposes a real-time, non-invasive Artificial Intelligence-based framework for automatic emotion recognition and stress level prediction using facial expressions. The system integrates image preprocessing, face detection, and a deep Convolutional Neural Network (CNN) for multi-class emotion classification. Based on detected emotional states, stress levels are inferred using an emotion-to-stress mapping model. Experimental evaluation on benchmark facial expression datasets demonstrates an emotion classification accuracy of 92% and stress prediction accuracy of 90%. The proposed approach offers a scalable and cost-effective solution for stress monitoring in education, workplace, and healthcare environments.