The cassava brown streak virus does not announce itself. By the time a farmer walking a field can see the characteristic leaf mottle and root necrosis, a significant portion of the crop is already unsalvageable. The window for effective intervention — fungicide application, varietal replacement, quarantine of affected rows — has typically already closed.
Shamba AI, developed by a Nairobi-based team and now operated in partnership with the Kenyan Ministry of Agriculture, gives farmers that window back. The system layers weekly Sentinel-2 satellite imagery over a national soil and climate database, flags anomalous spectral signatures at plot level, and pushes an alert to a farmer's feature phone or smartphone with plain-language guidance in Swahili, Kikuyu, Luo, or Kalenjin.
Where the satellite signal is ambiguous, farmers can photograph affected plants directly. The on-device model — compressed to under 40 megabytes to function on low-specification Android handsets — returns a diagnosis and treatment recommendation within eight seconds, without requiring internet connectivity at the moment of use.
A three-season evaluation covering 2.3 million registered users across seven counties found that early-alert farmers caught disease outbreaks an average of 22 days earlier than a matched control group using conventional extension services, and suffered 31 percent lower yield losses. The system currently monitors cassava, maize, tea, and coffee, with bean and sorghum integration scheduled for the 2026 long-rains season.
"Extension officers cannot be everywhere," said Dr. Grace Wanjiku, Shamba AI's lead agronomist. "A satellite can."