This article introduces EfficientNet, a smaller and faster model ideal for quick-result applications in detecting early signs of Diabetic Retinopathy. It explains how EfficientNet operates using Neural Architecture Search (NAS) to establish a baseline network and then undergoes compound scaling. The Messidor dataset is used to train the EfficientNet model, which can classify and identify the severity of diabetic retinopathy based on visible lesions in the retinal images. EfficientNet's performance compares favorably to other state-of-the-art models due to its unique scaling methodology, offering better accuracy with fewer parameters and less computational complexity.