Someone we know farms millet outside Tahoua. He had a question about a water dispute in his village — the kind of question where the answer depends entirely on which rule applies and who administers it.
He asked in Hausa. He got an answer in French, citing a legal code from the wrong country.
It is easy to file that under amusing model failure. It is not. He asked a real question, in the language he thinks in, and the system he reached was not built with him in mind at any point in its construction. Not in the data it learned from, not in the benchmarks it was measured against, not in the interface he typed into.
The gap is not talent
More than seventy million people speak Hausa. The language has centuries of written scholarship. What it does not have is presence in the datasets, models and tools that define what modern AI can do.
That absence compounds. No data means weak models. Weak models mean no usable tools. No tools mean no usage — and no usage means no new data. The loop closes on itself, and it stays closed until someone deliberately breaks it.
That is what Namu is for. Not a translation layer over someone else's system. Infrastructure built from the language outward.
