Project Gecko: Microsoft’s Blueprint for Truly Inclusive AI, Starting in Kenya
Microsoft has launched Project Gecko, an ambitious global research initiative headquartered in Kenya, fundamentally aimed at redesigning Artificial Intelligence to be truly inclusive of communities currently underrepresented in large language models (LLMs). This project is a decisive pivot away from simply adapting existing, predominantly English-centric AI.
The collaboration spans the globe, led by expertise from Microsoft Research Africa (Nairobi), Microsoft Research India, and the Microsoft Research Accelerator in the U.S., uniting resources to build AI systems centered on the unique needs of the world’s majority.
Project Gecko’s innovation relies on a trio of technological shifts designed for efficiency and local relevance:
- MultiModal Critical Thinking Agent (MMCTAgent)
Central to the initiative is the MultiModal Critical Thinking Agent (MMCTAgent). This AI system is engineered for sophisticated cross-modal reasoning, enabling it to analyze and synthesize information from speech, images, and video simultaneously.
Unlike traditional text-heavy systems, the MMCTAgent generates locally grounded, actionable answers using multiple communication formats tailored to the user’s need—a crucial feature for communities relying on oral knowledge and visual instruction.
2. Efficiency through Small Language Models (SLMs)
To ensure operational viability in challenging environments, Project Gecko utilizes Small Language Models (SLMs). This focus is a practical necessity, enabling the AI to run cost-effectively and efficiently on the low-cost, low-bandwidth devices prevalent in rural areas. This approach flips the industry trend of ever-larger models, prioritizing speed and accessibility over sheer parameter count for greater real-world impact.
3. Native Language Data Synthesis
A major technical hurdle—the significant deficit of training data for African languages—was overcome by building foundational tools from scratch. The team developed new automatic speech recognition (ASR) and text-to-speech (TTS) engines optimized for African phonetics.
The initiative currently supports widely spoken Kenyan languages, including Swahili, Kikuyu, Kalenjin, Dholuo, Maa, and Somali, powered by a crowdsourced dataset of over 3,000 hours of Kenyan speech. This local data foundation is key to generating culturally and linguistically authentic responses.
The initial deployment focuses on agriculture in Kenya, a sector vital for millions of smallholder farmers who depend on hyperlocal, community-based knowledge.
Project Gecko directly enhances tools like Digital Green’s FarmerChat, a speech-first assistant. For example, a farmer speaking Kikuyu in Nyeri County can verbally ask a complex question and receive a context-rich, multimodal response including:
- Text instruction.
- Audio explanation.
- A relevant video clip that auto-jumps to the precise instruction point.
Early field studies in both Kenya and India have decisively shown that for the global majority, local relevance and cultural grounding are stronger drivers of usability, trust, and impact than the sheer size or generic capability of the model.
Microsoft plans to expand Project Gecko into other critical sectors requiring deep local context, including healthcare, education, and retail, solidifying a commitment to ensuring the next generation of AI is shaped by the communities it is built to serve.