Generative AI is reshaping productivity and access to information around the world. However, these systems often fail to deliver the same quality of results for communities that are under-represented online—especially populations that speak low-resource languages. Because most AI models are trained on dominant, internet-rich languages, they frequently struggle to understand the linguistic, social, and cultural realities of the global majority.

This imbalance has led to low adoption of AI technologies in regions where local languages dominate. Even with improving connectivity and rising smartphone penetration, many communities still receive inaccurate or incomplete AI responses.

Microsoft’s Project Gecko is designed to close this gap.
Led by Microsoft Research, the initiative aims to build cost-effective, adaptable AI systems that work for everyone, regardless of language, culture, or available infrastructure.

What Is Project Gecko?

Project Gecko brings together experts from:

• Microsoft Research Africa (Nairobi)

  Ad

• Microsoft Research India

• Microsoft Research Accelerator (United States)

• Digital Green

• Partners across agriculture, academia, and philanthropy

Their mission: create AI systems that speak local languages, reflect community knowledge, and work through text, speech, and video—even on low-cost devices.

At the center of this initiative is the MultiModal Critical Thinking Agent (MMCTAgent), a new AI system capable of understanding speech, images, and video and providing responses grounded in local context. MMCTAgent is now available on Azure AI Foundry Labs, with open-source code on GitHub.

As Ashley Llorens, Corporate Vice President and Managing Director of the Microsoft Research Accelerator, says:

“Building AI systems from the ground up shaped by the knowledge, languages, and modalities of the global majority yields more innovative, useful solutions for a great number of people.”

Why Agriculture Is the Starting Point

Although Project Gecko will expand into healthcare, education, and retail, Microsoft chose agriculture first—and for good reason.

In countries like Kenya and India, agriculture is a major economic pillar, employing millions of smallholder farmers. Yet the sector faces unique challenges:

• Multiple local languages spoken in a single region

• Cultural nuances that vary from district to district

• Dependence on oral instruction and visual demonstrations

• Limited bandwidth and low-capacity devices in rural areas

Existing AI-powered agricultural apps often provide generic or incorrect information because their models weren’t trained on localized data.

Tanuja Ganu, Director of Research Engineering at Microsoft India, explains:

“Agriculture has very specific terms, which may change from language to language, and even district to district. All those domain-specific nuances need to be understood.”

To address this, Project Gecko builds on Digital Green’s FarmerChat, a speech-first tool used by agricultural extension workers. Digital Green’s library includes 10,000+ farmer-generated videos in more than 40 languages, which historically were difficult to access due to linguistic limitations.

With Project Gecko, this wealth of knowledge becomes searchable, understandable, and actionable.

How MMCTAgent Improves Accuracy and Trust

MMCTAgent enhances traditional AI by combining reasoning across:

• Audio

• Visuals

• Text

The agent breaks down complex questions, validates its own answers, and grounds its responses in real agricultural practices captured in community-generated content.

For example, a farmer in Nyeri County, Kenya can now ask a question in Kikuyu and receive the answer:

• in text

• in audio

• in video—jumping straight to the timestamp that demonstrates the solution

Field studies in Kenya and India show major improvements in accuracy, usability, and trust.

Lakshmi Devi, a farmer from Bihar, India, shares:

“Before, we would ask neighbors or dealers for advice and weren’t sure it was right. With FarmerChat, we ask our questions, follow the instructions, and see better results.”

Building AI Models for Under-Resourced Languages

A key finding from Microsoft’s research is that farmers prefer voice interactions, yet many local languages lack basic tools for:

• Automatic Speech Recognition (ASR)

• Text-to-Speech (TTS)

• Machine Translation

Existing datasets were too small or incomplete to train effective models.

To solve this, Project Gecko is building these tools from scratch, using small language models (SLMs) optimized for low-cost devices. These models often outperform larger LLMs in specialized domains like agriculture.

Key milestones:

• Support added for Swahili, Kikuyu, Kalenjin, Dholuo, Maa, and Somali

• Creation of a 3,000-hour Kenyan speech dataset

• A public leaderboard underway to benchmark African language performance

• Co-creation with more than 130 farmers to shape features like clarifying questions and actionable recommendations

Looking Ahead: Scaling Inclusive AI Worldwide

Project Gecko represents Microsoft’s long-term commitment to building AI that is:

• Locally adaptable

• Culturally relevant

• Equitable

• Capable of running on limited infrastructure

The team is developing a multilingual playbook to guide developers in building domain-specific, locally grounded AI systems.

By starting in agriculture and studying real-world use cases, Microsoft aims to create patterns that can be reused in sectors such as:

• Healthcare

• Education

• Retail

• Government services

As Tanuja Ganu notes:

“Our goal is to ensure that the next generation of AI is not only powerful, but also globally inclusive, culturally relevant, and shaped by the communities it aims to serve.”