From leaf photo to treatment plan in seconds.
Five steps, no lab visit. Here is exactly what happens between tapping upload and reading your answer. Models, data, and all.
STEP 01
Snap a leaf
Upload up to four photos from chat, the home prompt, or the landing hero. Images are compressed in your browser first, so sends stay fast even on slow connections. Then each file travels once to the server.
STEP 02
Get a diagnosis
A ResNet-50 classifier, trained on 38 crop disease classes from PlantVillage, reads every photo at 224×224 pixels and returns a diagnosis with a confidence score plus the top three predictions. Each photo is predicted independently, so mixed uploads stay separated.
STEP 03
Understand what it means
Every prediction is enriched from a 38-entry knowledge base into causes, treatment, and prevention. These are the same facts that fill your result card. Healthy leaves are always marked low severity, never alarming.
STEP 04
Talk it through
The assistant replies over a streaming language-model API, grounding every answer in your diagnosis context. With several photos it labels each one (Image 1, Image 2, …) and compares them. Ask your questions in plain words; the reply types out live.
STEP 05
Private by design
Your leaf photos are never sent to the language model. Only diagnosis text is shared. Temporary chats are processed and discarded, never stored. Review sessions, use two factor authentication, and delete anything, anytime.
Try it on your own crop
One clear leaf photo is all it takes.
