A retinal screening platform that

A retinal screening platform that classifies Glaucoma and Diabetic Retinopathy from fundus images, built so a clinic can use it rather than so a notebook can run it. Transfer learning on MobileNetV2 and EfficientNetB3 gave accurate classification at a model size small enough to serve without GPU inference costs. The harder part was everything around the model: upload and validation pipeline, patient records, scan history, a results dashboard with confidence scores, and role-based access. Stack: Python, TensorFlow, OpenCV, REST API, web dashboard.