Diabetic retinopathy is a prevalent disease that causes vision impairment. Treatment should be taken to avoid such tragic circumstances. This paper discusses an AI focused diabetic retinopathy system which aims to improve both accuracy and XAI transparency during prediction stage. A Resnet50 model is used in the classification of the DR severity levels by capturing the retinal images. Model ambiguity is lowered by using Grad-CAM, which illuminates areas of the retina that are critical to the model’s decision. The system automates image preprocessing, feature extraction and classification. The system decreases the human intervention, thus increasing efficiency.