Microsoft Research has launched Project Gecko, an initiative aimed at building AI systems that work reliably in low-resource and oral-first communities, with Kenya serving as one of the project’s central testing grounds.
- •The effort marks a shift away from English-dominant AI development toward models that function natively in Kenyan languages and rural contexts where speech, local dialects, and community media drive everyday communication.
- •Led jointly by Microsoft Research Africa in Nairobi, Microsoft Research India, and the global Accelerator group, Gecko focuses on creating small and large language models tailored for Kiswahili, Kikuyu, and other languages that rarely feature in mainstream AI datasets.
- •The team is building speech-first interfaces designed to work in villages and low-connectivity regions, where text-heavy digital services often fall short.
Agriculture is the project’s first major deployment area. Microsoft researchers are stress-testing their models through Farmer.Chat, a Digital Green web app used by smallholder farmers to ask farming questions through voice.
Instead of relying on generic online sources, the system draws guidance from community-generated agricultural videos, a grounding method intended to align AI responses with local farming practices and locally trusted knowledge. Early piloting initiatives in Kenya show improvements in answer quality, user trust, and ease of use.
The initiative runs on VeLLM, a platform created by Microsoft Research India to support multilingual, multimodal copilots and streamline localization for under-represented languages. Project Gecko is the first large-scale attempt to apply this platform in a real-world Kenyan setting, testing whether it can adapt AI outputs to local culture, accents, and oral knowledge systems.
If the approach holds, Gecko could lay the groundwork for AI copilots that serve Kenyan users at scale, not just in agriculture but eventually in education and other basic services.
The initiative is positioned as one of the clearest tests of whether large AI players can design technology around the languages, media habits, and lived realities of African users rather than retrofitting Western-trained systems for local use.
