Voice AI for Agricultural Advisory in Kenya under the Italy-India-Kenya Trilateral

Smallholder farmers reached through agricultural booking agents face four overlapping constraints: mobile coverage is intermittent and can weaken away from towns; prepaid data competes with other household spending; devices are often feature phones or low-end Android handsets; and literacy may be limited in the languages offered by digital interfaces. Extension officers cannot provide individual advice to thousands of farmers on demand. Text-first services leave the literacy barrier unresolved, and internet-dependent services depend on affordable data, suitable devices and usable connectivity. SMS and USSD do not need internet access or a smartphone but still require mobile coverage and airtime. Before this deployment, booking agents handled advisory manually by phone and SMS, and participating hubs had no local-language voice advisory and no hub-level weather advisory in Kiswahili. The pathway combines familiar-language voice interaction with weather advice delivered through existing agricultural relationships, aiming for usable agricultural support within farmers' device, language and spending constraints.

Crane AI Labs builds offline-first voice and language AI for low-resource African languages, with delivery on low-cost handsets across health, agriculture and education. The organisation operates in Uganda, Kenya and Ethiopia. Offline means inference on the device without a network connection; network-delivered services are described separately. Crane is the implementing and orchestrating partner for the Voice AI Pathway under the UNDP AI Hub for Sustainable Development and the Italy-India-Kenya trilateral. It connects speech models, compute, delivery channels, enterprise partners and evaluation into a working deployment. Crane AI Labs is the first African team in Google DeepMind's Gemma partner programme. Its Gemma work supports familiar-language AI on constrained devices. Its Luganda model work uses CINECA's Leonardo supercomputer through the AI Hub. The Uganda Cultural Context Benchmark (UCCB) is included in the UK AI Security Institute-linked Inspect Evals repository.