The obvious move for an African language AI company is to announce support for as many languages as possible. It reads well and it is largely meaningless, because supporting a language and serving it are different claims.
We are starting with Hausa, and we are going to stay there until the foundation is genuinely solid.
Why one language first
Everything that makes a language model useful is language-specific and slow to build. Evaluation sets that measure the right things. Dialect coverage that reflects how people actually speak. Domain vocabulary for the sectors where the tool has to work. None of that transfers for free.
Getting it right once also teaches you the method. The second language is faster because you know what to collect, how to evaluate it, and where the failure modes hide — not because the model weights carry over.
What comes after
Zarma is next, spoken widely along the Niger river in the southwest. Then Fulfulde, which runs in a band across the Sahel from Senegal to Cameroon and has always crossed borders and dialects.
Both are chosen for the same reason Hausa was: large communities, real daily use, and almost nothing built for them.