Why Our AI Answers in Swahili

Most AI products on the market today were built with an unspoken assumption: that the user thinks, types, and reads in English. The interface is in English. The model is strongest in English. The examples in the documentation are in English.

For a large part of the world, that assumption is invisible because it happens to be true for them. For a large part of Tanzania, it is a closed door.

We think that door should be open.

Language is not a feature. It is the whole interaction.

When a tool only works well in your second or third language, you do not get a slightly worse experience. You get a fundamentally different one. You hesitate. You simplify what you were going to ask. You second-guess whether the answer means what you think it means. The friction is not in the software. It is in your head, the whole time you are using it.

A student trying to understand fractions does not need an AI that can also write them a sonnet. They need one that explains the idea clearly, in the language they think in, using examples from a world they recognise.

A tutor that teaches in Swahili, with examples from here

Inside Ubunifu Insight, one of our AI agents is an Education Tutor designed for Tanzanian students following the national curriculum. It does not just translate English explanations into Swahili. It teaches in Swahili, the way a good teacher here would.

Ask it to explain fractions and it does not reach for a pizza. It reaches for a chapati: one whole chapati cut into four equal pieces, each one a quarter. It connects the idea to cooking, to farming, to splitting things fairly: the everyday situations where fractions actually show up in a Tanzanian student's life.

That is a small thing and an enormous thing at the same time. The maths is identical. But the explanation lands, because it starts from something the student already understands instead of something they have to decode first.

Why this matters beyond education

The tutor is one example, but the principle runs through everything we build. The businesses and people we build for do not operate in a single language, and they should not have to switch into someone else's defaults to use good software.

That means:

• AI that can read a document and answer questions about it in the language the question was asked in.
• Tools that use local examples, local context, and local realities instead of imported ones.
• Software that meets people where they are, rather than asking them to meet it halfway.

This is not about being clever with localisation settings. It is about who the software is actually for. A tool built for "global users" in practice tends to mean a tool built for English-speaking, well-connected, high-income users, and then offered to everyone else as-is.

The opportunity in the gap

Here is the part that should interest anyone building software: this gap is not a charity case. It is an opportunity. There are millions of people for whom the current generation of AI tools is technically available and practically unusable, purely because of the language and context they were built around.

Build for those people properly, in their language and their context, and you are not making a worse product for a smaller market. You are making a better product for a market almost nobody is serving well.

We answer in Swahili because the people we build for think in Swahili. It really is that simple, and almost nobody is doing it.