Building software for Kinyarwanda speakers

Most of the tools I use every day assume English. Some of my side projects don't get to.

A few of them are built for Kinyarwanda speakers: people who watch, search and learn in Kinyarwanda first.

That changes the work in ways you don't notice until you're in it.

The defaults don't fit

Fonts, date formats, payment providers, voice models, even the copy a designer drops in as placeholder. Everything comes tuned for someone else.

So a lot of the job is noticing defaults and asking whether they make sense for the person on the other side of the screen. Usually they don't, quite.

Voice is the hard part

I've been testing Kinyarwanda text-to-speech. The surprising part: the model says every word well. What it gets wrong is the pausing. Where a sentence breathes, where it stops.

That's a good lesson in general. Being technically correct, word by word, isn't the same as sounding right. Users feel the rhythm before they notice the words.

Small market, real problems

It's tempting to build only for the biggest market. But a smaller audience that nobody is building for is a better place to be useful. People notice when something is made for them.

What I've learned

• Test with real speakers early. They hear problems you can't.
• Don't translate the interface last. Write it in the language from the start.
• Local payment methods aren't a nice-to-have. They're the product.

Building for people who are usually an afterthought is some of the most satisfying work I do.