DRAGS is Bangladesh's first clinical-grade retinal AI system: a deep learning model trained on 63,077 NHS DESP-graded fundus images to automatically classify diabetic retinopathy severity and detect lesions, built specifically for mass triage where specialists are scarce. Achieving 94.2% grading accuracy and a mean lesion detection AUC of 0.981 across 10 pathologies, the system delivers immediate, actionable referral decisions aligned to NHS DESP standards, not just descriptive labels. Deployed via RetinaBuddy, a smartphone app requiring no specialist equipment, it has already powered Bangladesh's first AI-driven free eye camp in the Chittagong Hill Tracts, bringing sight-saving screening to tribal communities for the first time.