Kookoo Doctor
Cocoa disease diagnosis for farmers, from a phone photo.
Overview
A farmer photographs a cocoa pod or leaf and gets a plain-language read on black pod disease or CSSVD - with an honest confidence level, what to do next, and the option to ask a follow-up question. Built at the Build with Gemma 4 AI Hackathon (July 2026).
What makes it work
- Refuses to guess: blurry, dark or off-subject photos are flagged for a retake instead of diagnosed
- Confidence is tied to visible structural evidence (lesion shape, texture, spread) - never to how dark a pod looks
- Actively rules out benign look-alikes such as ripening colour change, dirt, sun-scorch and old insect marks
- Every farmer-facing sentence is rewritten in everyday language - no “lesion”, “necrosis” or “sporulation”
- Destructive advice (remove and burn a pod) is always conditional on a second look or an extension officer
- Cost-of-inaction calculator uses documented per-farm loss ranges and refuses to invent numbers it cannot back
The story
Cocoa is Ghana's biggest export crop, and black pod disease alone can wipe out most of a farm's yield in a bad season. Extension officers from CRIG cannot be everywhere, so a farmer who sees a dark patch on a pod often has to guess.
Kookoo Doctor lets the farmer take a photo and ask. The model is Gemma 4, but the work is in the prompt: the system is told to assess image quality first, to look for structural evidence rather than colour, to name the benign explanation it ruled out, and to keep its confidence low when all it has is a dark patch. An uncertain “healthy” is downgraded to “keep watching” rather than false reassurance.
The output is rewritten for someone who may have limited literacy - short, spoken-language sentences - and any advice to destroy a pod is conditional on confirmation. The farmer can then ask a follow-up in the same context.
You can try it at kookoo-doctor.vercel.app - open it on a phone, point the camera at a pod, and read the answer the way a farmer would.