A goat coughs in rural Tanzania. The nearest vet is hours away, and the farmer cannot tell whether it is a cold or Peste des Petits Ruminants, a disease that can wipe out an entire herd within weeks.

That uncertainty is expensive. PPR is associated with estimated global economic losses of up to US$2.1 billion annually, and the people it hurts most are smallholder farmers with the least access to veterinary care.

For my final-year project, I asked a simple question: what if a phone could answer before the vet arrives? I trained an XGBoost model on 21,199 clinical records from Tanzania, using nine observable symptoms that any farmer or field officer can check without laboratory equipment. The model reached 80.61% accuracy, and my Flask application returned diagnoses in under two seconds during testing.

The biggest lesson was not technical. Good AI for agriculture is not about the fanciest algorithm; it is about speed, accessibility, and meeting people where they are. A model that needs a lab is useless in the field. A model that answers in two seconds, on a phone, in plain language, can change what happens next. Sometimes, saving an entire herd starts with a two-second answer.