Diabetic retinopathy is among the increasing causes of loss of vision in the population with diabetes worldwide. Early diagnosis with determination of the severity level of retinopathy is very important for proper management and treatment of the disease. In this research work, an AI model for automatic detection of various severity levels of retinopathy using deep convolutional neural networks (CNNs) developed on the Tensor Flow platform is suggested. This suggested model utilizes deep learning methods so as to classify various levels of severity of diabetic retinopathy including no retinopathy, mild, moderate, severe and proliferative retinopathy. The model was also tested and trained on a large database of retinal images and achieved high accuracy in detection and measuring the severity of retinopathy-causing lesions. This kind of automated systems would assist medical professionals in early detection and monitoring of diabetic retinopathy, which would ease the workload on healthcare providers as the system would be an automated system.