LeafScan AI
A web app that looks at a photo of a plant leaf and tells you exactly what disease it has - or if it's healthy. Drag in a photo, get the top 3 diagnoses in seconds.
What it is
LeafScan AI is a web app that can look at a photo of a plant leaf and tell you exactly what disease it has - or if it's healthy. You drag in a photo, it runs through a model I trained, and within seconds you get the top 3 most likely diagnoses with confidence scores. No technical knowledge needed, works on any device.
Why it exists
Most people can't tell what's wrong with their plants until it's already too late. Professional diagnosis isn't accessible to regular farmers or gardeners. LeafScan makes it as easy as taking a photo.
How it works
Drag a photo into the web interface - works from a phone, laptop, or tablet. No account, no install.
The image is resized and normalized so it matches the format the model was trained on.
EfficientNetB0 runs over the image and scores it against all 38 disease classes in one pass.
You get the three most likely diagnoses, each with a confidence score and a healthy / diseased label.
Inference, visualized
A loop that mirrors the live app - a leaf comes in, gets preprocessed, the model scores it against all 38 classes, and the top 3 predictions come back with confidence scores.
What it can do
Challenges & learnings
Where it fits
Farmers can catch outbreaks from a phone photo before the damage spreads through a field.
A fast classifier for field studies - researchers can label samples in seconds instead of sending them to a lab.
Same model can plug into drone-based crop scanning or automated greenhouse monitoring systems.
Tech stack
Model trained in Google Colab on GPU, served via Flask on Render as a REST API. Frontend in plain HTML/CSS/JS deployed on Vercel - the two talk over a simple base64-encoded image endpoint.