Original article Total Telecom:Read More
News
Microsoft has unveiled Project Gecko, a research-driven initiative aimed at producing cheaper, localised generative AI systems for populations under‑represented in current models , starting with smallholder farmers in Kenya and India.
The project, led by Microsoft Research with contributions from Microsoft Research Africa in Nairobi, Microsoft Research India, the Microsoft Research Accelerator in the US and partners including agri‑tech NGO Digital Green, seeks to tackle the language, cultural and infrastructure barriers that limit AI uptake in low‑resource settings.
At the centre of the effort is the MultiModal Critical Thinking Agent (MMCTAgent), a multimodal system that ingests speech, images, and video to produce context‑rich, locally grounded answers. Microsoft says MMCTAgent can break complex queries into sub‑questions, verify its own outputs, and anchor responses in community‑generated practice captured in videos and transcripts.
The system is available on Azure AI Foundry Labs and its code has been published on GitHub.
Agriculture is Project Gecko’s first focus because of its economic weight in countries such as Kenya and India, where millions of smallholders work plots of under five acres. Microsoft and partners argue that existing AI tools often fail farmers because models are trained predominantly on English data, do not handle local dialects, and do not reflect region‑specific agronomic terms or practices. Farmers commonly rely on oral instruction and video demonstrations, both of which are channels that conventional text‑centric models struggle to exploit.
Project Gecko builds on Digital Green’s FarmerChat platform, which already serves millions of farmers and holds more than 10,000 agricultural videos in over 40 languages and dialects. Microsoft says Project Gecko enables a farmer in Nyeri County, for example, to ask a question verbally in Kikuyu and receive a text, audio, or video response, including a jump to the precise timestamp in a training clip. Field studies in Kenya and India reportedly show improved accuracy, usability and trust compared with generic AI systems.
A key technical strand of the work is creating speech infrastructure for under‑served languages. The team has collected roughly 3,000 hours of crowd‑sourced Kenyan speech and expanded support to Swahili, Kikuyu, Kalenjin, Dholuo, Maa, and Somali. To run on the low‑cost devices typical in rural areas, the project uses small language models (SLMs) and is preparing a public leaderboard to benchmark African language performance.
Microsoft plans to broaden Project Gecko beyond agriculture into healthcare, education, and retail, and will publish a multilingual playbook for developers.
While major challenges still remain for the project, including limited connectivity for the target consumers and sparse datasets for many languages, Project Gecko represents a significant attempt to bridge the rapidly growing digital divide.
Also in the news
Connected Britain Award winners 2025 announced!
Netomnia announces ‘powerful and ambitious’ rebrand ahead of Connected Britain
VodafoneThree drops Samsung, relies on Nokia and Ericsson for £2bn network upgrade