Ikawa

Ikawa is an offline "yield detective" for smallholder coffee farmers, built for the Agriculture track of the Hack-Nation × World Bank Small AI for Development hackathon. It runs on a low-end Android phone with no signal. A farmer photographs leaves from a bad row and a good row, answers six picture questions by voice or tap, and gets a ranked list of 9 likely causes of her yield drop, plus one action card played as audio in Kiswahili. When the evidence is thin, it says so and sends the case as one SMS to the extension officer, so a person makes the final call. Voice answers are matched against the farmer's own recordings (MFCC and dynamic time warping), which needs no speech model. Each language is a pack, so new languages need no app code changes. The leaf model is 2.4 MB (int8 ONNX), and the whole app is about 17 MB and works in airplane mode.