Diabetic Retinopathy (DR) is a progressive eye disease commonly affecting individuals with diabetes mellitus, and it remains a leading cause of preventable blindness among adults worldwide. Characterized by retinal damage, abnormal growth of fragile blood vessels, and bulging of retinal blood vessels, DR reflects the systemic impact of prolonged diabetes. Diagnosing DR typically involves a comprehensive eye examination conducted by an ophthalmologist, including visual acuity tests, pupil dilation, and tonometry to identify key symptoms. Recent advancements in DR classification have leveraged innovative Artificial Intelligence (AI)-based techniques to address challenges such as noisy images, class imbalances, and inconsistent resolutions. This research proposes a novel Convolutional Neural Network (CNN) architecture incorporating a Channel Spatial Focus Block (CSFB) attention mechanism. The model is designed to perform both binary and multi-class classification of DR, enabling early detection of microaneurysms and other critical DR symptoms, which is essential for controlling disease progression. The model was trained using the APTOS 2019 and DDR datasets. Our proposed approach achieved an accuracy of 99.32% for binary classification and 84.99% for multi-class classification on the APTOS 2019 dataset. On the DDR dataset, the model achieved accuracies of 90.67% for binary classification and 84.16% for multi-class classification. These results were benchmarked against state-of-the-art transfer learning models, demonstrating the effectiveness of our method. Further, reinforcement learning is a new and emerging paradigm in AI which uses agent&ndashenvironment methods to produce the results. Where an agent is optimally trained to do the classification task. The completion of the work will bring a complete automated tool for early screening/ diagnosis of Diabetic Retinopathy.