Diabetic retinopathy (DR) is a leading cause of vision impairment and blindness among individuals with diabetes, making early diagnosis crucial to prevent severe complications. Traditional screening methods are often invasive and costly, limiting accessibility. This introduces an innovative, non-invasive approach by leveraging pupillometry, which analyzes pupil response to light stimuli for DR detection. This system integrates an ensemble of deep learning models, including convolutional neural networks (CNNs) and transformer-based architectures, to classify DR across five severity stages: Mild, Moderate, Severe, Proliferative, and Early DR. This project is deployed on the streamlit platform, allowing clinicians to upload pupillometry data and receive real-time diagnostic results with high accuracy and confidence. By combining advanced artificial intelligence with a cost-effective, scalable solution, this project enhances early DR detection and expands global accessibility to timely treatment and intervention.