-Developed a multi-disease retinal classification system on 20,000+ fundus images detecting 6 eye diseases using an ensemble of Vision Transformer and binary CNNs, achieving 93.7% accuracy. -Applied knowledge distillation to compress the ensemble into a lightweight CNN, retaining 82.7% accuracy for efficient edge deployment. -Built end-to-end data pipelines with preprocessing and augmentation techniques to improve model robustness across diverse real-world retinal imaging conditions.