Diabetic retinopathy is a critical public health problem and one of the top reasons of blindness among adults. As such, early detection and accurate classification are very important in preventing the progression of disease and, hence, the load on health systems. In this study, we propose an ensemble model with InceptionResNetV2, InceptionV3, and ResNet-50 architectures to detect and classify diabetic retinopathy using the APTOS 2019 Blindness Detection Dataset. Our ensemble model borrows the strengths from these three models in an attempt at attaining class labeling accuracy for retinal images regarding the different severity levels of Diabetic Retinopathy. In this approach, individual model predictions are ensemble to improve overall performance, especially when certain models are weak. The work essentially leads to the development of automated tools for blindness detection, with solemn promise for rendering early diagnosis and treatment for diabetic retinopathy with resource constraints.