In 2024, Nigeria quietly embarked on one of the most ambitious technology projects in its history: building a homegrown, open-source large language model (LLM) capable of understanding the country’s diverse languages, dialects, and accents. The initiative, developed in partnership with Lagos-based frontier tech company Awarri, is named N-ATLAS, and represents an attempt to create an AI system truly rooted in the daily linguistic realities of Nigerians.One year later, on September 21, 2025, at the 80th United Nations General Assembly (UNGA80) in New York, Minister of Communications, Innovation, and Digital Economy Bosun Tijani elevated N-ATLAS from a technical experiment to a national mission.> “This is a commitment to unity, inclusion, and global contribution,” Tijani declared. “N-ATLAS places Nigerian—and African—voices at the centre of the digital future.”—Awarri’s Evolution: From Robotics to Frontier AIAwarri did not begin as an AI lab. Its co-founder, Silas Adekunle, gained global recognition years earlier with Mekamon, the groundbreaking consumer robot that became the first of its kind sold in Apple Stores. Today, the company describes itself as a “360 AI organisation,” spanning robotics, data engineering, and model development.For Itua Aizehi, Awarri’s Vice President of Marketing and Communications, the company’s mission is simple:> “Africans can solve African problems through technology. To do that, we must build foundational tools—not just consume what others create.”The name Awarri, derived from the Yoruba word awari (“to seek and find”), reflects the company’s obsession with discovery and invention.—N-ATLAS: Building an AI Trained on Nigerian VoicesUnlike global AI models trained predominantly on English, Mandarin, and a handful of European languages, N-ATLAS aims to reflect the linguistic complexity of Nigeria.The backbone of the project is data—thousands of hours of it. With support from the government’s 3 Million Technical Talent (3MTT) programme, Awarri built LangEasy.ai, a platform for collecting voice samples from everyday Nigerians. Contributors recorded phrases in:YorubaIgboHausaEfikIbibioNigerian PidginAccented Nigerian EnglishThese recordings were cleaned, transcribed, and annotated before feeding into the model.Awarri has already released early versions supporting Yoruba, Igbo, Hausa, Nigerian-accented English, and Pidgin—with accuracy levels reportedly above 80%.For Awarri lead engineer Sunday Afariogun, the voice-first approach is vital:> “Many Nigerians can speak their languages but do not write them fluently. A farmer should be able to ask a question in Hausa and get an immediate answer—without English, and without literacy barriers.”—Why Not Just Use ChatGPT or Gemini?Some critics argue that Nigeria does not need its own LLM. But Awarri’s engineers counter that Western models simply don’t reflect Nigerian speech patterns, cultural references, or local context.> “ChatGPT or Gemini can attempt Yoruba or Pidgin, but not the way Nigerians speak them,” Afariogun explains. “If our languages aren’t represented in AI, we risk losing them.”Aizehi frames it more sharply:> “This is about cultural sovereignty. Out of over 7,000 languages globally, fewer than 30 appear in AI datasets. Africa has over 2,000 languages, but less than 2% are digitised. If we don’t preserve them, no one will.”—The Government’s Role — and the Biggest BottleneckThe federal government has been instrumental in rallying contributors, providing institutional backing, and signalling interest in supporting AI infrastructure. But compute power remains the project’s biggest obstacle.Training advanced AI models requires enormous GPU resources—hardware Nigeria does not yet have at scale.> “We’re nowhere near ChatGPT’s capacity,” Afariogun admits. “We rely on Amazon and Google cloud services because we don’t have local data centres built for large-scale AI.”—Open Source as a Strategic ChoiceUnlike global tech giants, which guard their models behind paywalls, Nigeria is taking a radically different approach by open-sourcing N-ATLAS.Aizehi explains the philosophy:> “We want a foundation the entire ecosystem can build on—startups, developers, researchers, even government agencies.”The hope is that an open ecosystem will spark innovation similar to what Linux and Android did for global software development.—Why N-ATLAS Matters — Real Use CasesDevelopers and entrepreneurs see massive potential:EducationBilesanmi Faruk, CTO of edtech company Lena, says N-ATLAS could enable low-cost classroom tools for rural students where internet connectivity is limited.AgricultureKrosAI co-founder Joshua Firima imagines farmers calling a number for advice in Tiv, Hausa or Yoruba-accented English.ResearchNLP researcher Zainab Tairu says the availability of indigenous language models could finally accelerate African AI research.—Challenges AheadDespite enthusiasm, several hurdles remain:High training costs due to reliance on foreign cloud platformsWeak local infrastructure with limited AI-ready data centresBroadband penetration below 50%, excluding millions from AI accessFunding gaps for research and expansionAfariogun says the plan is to start with early adopters—developers, startups, and urban users—and gradually expand to underserved communities.—A Future Where AI Speaks Like NigeriansIn the long run, the success of N-ATLAS will not be measured by accuracy scores alone but by whether Nigerians embrace it in daily life.For Aizehi, the mission remains urgent and non-negotiable:> “If we don’t build AI for ourselves, nobody will. This is about preserving our identity and ensuring that when AI speaks, it speaks like us.
5 / 5
