AnkitSingh1827/Retinal-Screening-AI ? reverse-engineered prompt

Build me a Streamlit web app for AI assisted retinal screening.

I want a user to upload a retinal fundus image, have the app check if the image is usable, then predict one of four results, Normal, Diabetes, Glaucoma, or AMD. Please show the main prediction, the confidence score, and the full class probabilities in a clean medical style interface. I also want a Grad CAM heatmap so it’s easy to see which part of the retina influenced the result.

Please include a clear warning that this is only a screening tool and not a medical diagnosis, and that a licensed ophthalmologist should review the result. If the model confidence is low, show a caution message. Use the trained model files and project assets if they exist, and set things up so the app can run locally without too much extra work. If anything is missing, create sensible placeholders and look up current docs online if you need to.