AI has shown significant breakthroughs in medical imaging tasks by automating the diagnosis of various diseases. One such disease is diabetic retinopathy (DR), which is a more common cause of sight loss in India and around the world. This paper highlights the power of AI in building an automated detection model for DR severity detection that can encounter cases of blindness by early detection. However, in the medical field, the success of AI-based models is heavily based on large-scale labeled medical images. Hence, in this work, IDRiD dataset is utilized to build a DR severity detection model using limited labeled fundus images. To further enhance the performance of the model, hyperparameter tuning and data augmentation are complemented during the building of the model. The experimental results show that the model attains an accuracy of 0.836 and a Cohen’s Kappa score of 0.711 on the task of predicting the severity of DR. However, it is realized that the supervised learning model struggles to provide optimal performance when subjected to a limited amount of labeled data. This highlights the requirement for annotation-efficient techniques to train models with limited labeled data, which is discussed in the future work section.