Nigerian AI startup ("awari" = "to seek and find" in Yoruba) that built N-ATLAS -- Nigeria's first government-backed open-source multilingual AI model suite, launched at UNGA80 in September 2025. N-ATLAS includes ASR models for Yoruba, Hausa, Igbo, and Nigerian-accented English (Whisper fine-tunes) plus a multilingual LLM (Llama-3 8B fine-tuned on ~392M tokens). Also operates LangEasy (crowdsourced language data collection platform targeting 500K hours), RootLens (visual data management), and a data annotation lab in Ikorodu employing 100+ workers. Leading Nigerian telcos and banks integrating N-ATLAS for customer service.
Building open-source Nigerian language models (LLM + ASR) with government backing. N-ATLAS is the most significant government-private partnership for African language AI. Voice-first approach aligns with low-literacy use cases. LangEasy's crowdsourced data collection model could be leveraged for UNICEF's community engagement. Already partnered with NCAIR (existing actor).
N-ATLAS models are published under the NCAIR1 HuggingFace org (government partnership), not under AWARRITech. Human evaluation shows English (4.21/5) and Hausa (3.98/5) perform well, Igbo (3.87/5) is decent, but Yoruba (2.69/5) still needs work. Developer challenge with NGN 1M prizes to encourage ecosystem building around N-ATLAS.