Diabetic retinopathy (DR) is a primary contributor to blindness among diabetics, emphasizing the need for early diagnosis to mitigate vision loss. Traditional automated systems, relying on image processing and machine learning techniques such as random forests, support vector machines (SVM), and logistic regression, often require manual feature extraction, leading to suboptimal accuracy and poor generalization. These methods struggle to effectively capture intricate patterns in retinal fundus images, limiting their reliability in real-world clinical applications. To address these challenges, a hybrid deep learning approach is introduced, combining vision transformers (ViTs) and convolutional neural networks (CNNs) for DR detection. CNNs efficiently extract local features, such as micro-aneurysms and hemorrhages, while ViTs capture long-range dependencies and positional relationships across the retina, enhancing overall feature representation. The proposed system achieves 98% accuracy, demonstrating superior performance over existing methods. By integrating CNN and ViT, the model provides a more accurate, scalable, and robust solution for DR detection.