Microsoft has unveiled “Project Gecko”, a research initiative aimed at restructuring generative AI to serve the “global majority,” beginning with the agricultural sectors in Kenya and India.
The project seeks to solve a persistent market failure: the inability of current large language models (LLMs) to function effectively in low-resource languages or reflect non-Western cultural nuances.
Led by Microsoft Research Africa and India, the initiative addresses the dominance of English-centric training data, which hinders AI adoption in developing nations, even among those with internet access.
To counter this, Project Gecko deploys Small Language Models (SLMs). These compact models require minimal computing power, making them viable for the low-bandwidth, low-cost devices prevalent in rural economies.
Central to the rollout is the MultiModal Critical Thinking Agent (MMCTAgent), now available on Azure AI Foundry Labs and GitHub. Unlike standard text-based models, MMCTAgent processes speech, images, and video simultaneously. This allows a farmer to ask a question verbally in a local dialect, such as Kikuyu or Swahili, and receive an answer synthesized from specific timestamps within instructional videos. To power this, Microsoft amassed a new dataset comprising 3,000 hours of crowd-sourced Kenyan speech.
Microsoft selected agriculture as the beachhead market due to its outsized impact on GDP in the pilot regions. The tech integrates with Digital Green, a development organization with a library of 10,000 agricultural videos. Previously, this content was difficult to search or access due to linguistic barriers; however, the new system enables Digital Green’s “FarmerChat” to provide locally grounded, verified advice rather than generic, often inaccurate AI-generated content.
While currently focused on agri-tech, Microsoft views Project Gecko as a blueprint for healthcare and education sectors in emerging markets. The company plans to release a multilingual playbook for developers, aiming to standardize how AI is localized for populations that have been historically underserved by Silicon Valley infrastructure.
