If left untreated, diabetic retinopathy (DR) can cause significant vision damage, so early and accurate diagnosis is essential. This study presents a new approach to improve the accuracy of DR analysis using a hybrid convolutional neural network (CNN) model. Leveraging the capabilities of the ResNet50 and InceptionV3 algorithms, the model attempts to extract complex features from fundus images—features that are important for early detection of drug resistance. The problem is detecting DR early when symptoms are mild, making automated methods inaccurate. Adding other models such as DenseNet and Xception will increase the accuracy to 97.2%. Additionally, an extension uses the Flask framework to create an easy-to-use frontend with validation, which simplifies user testing. This comprehensive approach not only promises a better classification of DR, but also highlights the importance of early diagnosis; thereby limiting the risks of vision loss associated with this disabling disease is reduced.