Announcing C-elo Labs: AI for African Languages
December 13, 2025
Launch Announcement
Founder & Lead Researcher
Mark Gatere
We're launching C-elo Labs to build AI that speaks African languages natively. Starting with Kikuyu — a tonal Bantu language spoken by over 8 million people in Kenya — we're building translation, voice chat, and text-to-speech models that work.
overview
Over 2,000 African languages are spoken across the continent, yet virtually none have meaningful AI support. While ChatGPT converses fluently in English, speakers of Kikuyu, Kamba, Luo, and hundreds of other languages are locked out of the AI revolution.
C-elo Labs exists to change this. We're building production AI models — not research prototypes — for African languages. Our approach: fine-tune state-of-the-art open-source models (TranslateGemma, MMS, Mimi, Llama) on curated African language datasets, and deploy them as accessible products.
We chose Kikuyu as our first language because it's underserved (zero AI support despite 8M+ speakers), linguistically challenging (tonal semantics, agglutinative morphology, code-switching with English and Swahili), and because our founder speaks it natively — enabling direct quality validation.
Our initial product suite targets healthcare, agriculture, and customer service — verticals where millions of Kenyans face language barriers when interacting with technology and institutions.
Interviewees:
Mark
Gatere
Founder & Lead Researcher
AI researcher and native Kikuyu speaker building language technology for underserved African language communities. Background in machine learning and NLP.
——Why does Africa need its own language AI?
Gatere: Because translation isn't conversation. If you ask Google Translate 'Wĩ mwega?' (How are you?), it gives you the English equivalent. We want AI that responds 'Niĩ ndĩ mwega' (I am fine) — that understands and speaks the language naturally. That requires training models specifically on African language data, not just layering translation on top of English AI.
——What will C-elo build first?
Gatere: Three things: a translation engine (English to Kikuyu), a text-to-speech system (Kikuyu text to spoken audio), and a voice chatbot (speak Kikuyu, hear Kikuyu). Translation is the most mature — we have 30,430 sentence pairs already curated. Voice is the most impactful — most Kikuyu speakers interact with technology through speech, not text.
——What datasets are you working with?
Gatere: We have access to the African Next Voices corpus — over 750 hours of Kikuyu audio — plus Google's WAXAL dataset with 9 hours of studio-quality TTS recordings. For text, we've compiled 30,430 English-Kikuyu translation pairs from multiple sources including agricultural and general domains. This combination of speech and text data is what makes multi-modal Kikuyu AI possible.
——Where do you see C-elo in two years?
Gatere: Serving real customers. A hospital in Central Kenya using our voice AI for patient intake in Kikuyu. A bank's IVR system that understands Kikuyu callers. An agriculture extension service that farmers can call and ask about crop diseases in their mother tongue. Then we expand — Kamba, Luo, Kalenjin, and eventually across the continent.
