A passenger's ride has just arrived. Before she gets on, a message appears on her phone, asking her to check one thing: does the number plate on the vehicle in front of her match the one in the app?
It's one of the most important safety checks we've ever built. If the plate doesn't match, the trip is cancelled on the spot.
Now imagine that message is written in a language she reads slowly, or not at all. She glances at it, doesn't quite follow, taps the big button to make it go away and climbs on.
The check was there. The code worked. And it protected no one.
That's the moment we keep coming back to whenever someone asks why a young company would bother launching in three languages at once. Our answer is short: a safety feature you can't read isn't a safety feature.
One platform, many languages
Twende is launching in the Democratic Republic of the Congo and Uganda, and growing toward the rest of East and Central Africa. On paper, that sounds like two countries. In practice, it's a map of overlapping languages.
In the DRC, French is the official language of schools, government and business. In Kinshasa, everyday life also happens in Lingala. In the east, Swahili is the language of the street and the market. In Uganda, English and Swahili are both official languages, and across East Africa, Swahili is the bridge between countries.
The same person might read official documents in French, chat with friends in Lingala and bargain at the market in Swahili. There's no single "default user" whose language we can pick and call it done.
So from the very first version, Twende's website has spoken French, English and Swahili, and our passenger app already speaks a fourth language: Lingala.
Why "we'll translate it later" never works
Every product team has heard it: launch in one language first, translate once things settle down. We understand the appeal. We also think it's one of the most expensive shortcuts in software.
Translation isn't something you add on top of a product. It runs through all of it:
- Every piece of text has to live somewhere other than the code, or someone has to go back and find it, screen by screen.
- URLs have to know which language they're in, so links can be shared and search engines can send people to the right version.
- Dates, numbers and money have to be written the way each language writes them.
- Content like articles and help pages has to exist separately in each language, each with its own address.
- Layouts have to survive sentences that are longer in one language than another.
Do it on day one and each of these is a small decision. Do it a year later and it's a rewrite of everything you've built, while the product keeps growing underneath you.
We know this because we've seen it in our own work. Wherever we built a screen in a single language "for now", that "for now" lasted longer than planned, and catching up is slower than doing it right would have been. Translation is an architecture decision, not a content task.
What "built in" means on our website
Here's what three languages from day one looks like in practice on twende-drive.com.
Every language has its own address. French, the most widely read language in our first market, is the default and lives at the root of the site. English lives under /en and Swahili under /sw. Every page tells search engines where its other versions are, including which one to show when no language is specified. Someone searching in Swahili lands on the Swahili page, not on a French one with a language menu.
All the text lives outside the code. Every heading, button and message on the site comes from three translation files, one per language. Those three files have exactly the same structure, line for line. When a sentence is added in one language, there's an empty space waiting for it in the other two.
Articles are written, not just translated. Each blog post exists separately in each language, with its own address in that language. "Fixed Pricing" is /en/blog/fixed-pricing, /blog/prix-fixe in French and /sw/blog/bei-isiyobadilika in Swahili. A Swahili reader sees a Swahili link, not an English one with a Swahili page behind it.
Dates speak the reader's language. An article published on 25 January reads "January 25, 2026" in English, "25 janvier 2026" in French and "25 Januari 2026" in Swahili.
Even the metadata knows where it is. When a Swahili page is shared on social media, it's tagged as Swahili from the Congo (sw_CD), not as a generic Swahili page. It's a small detail, and that's the point. Respecting a language means getting the small details right too.
The bugs only multilingual products have
Building in several languages also means meeting bugs that single-language products never see. Two of ours are worth sharing, because they'll save someone else an afternoon.
The article that existed and couldn't be found
French has accents, so some of our French articles used to have accented addresses, like la-sécurité-avant-tout. For a while, some of them returned a "page not found" error, even though the article was sitting right there.
The cause turned out to be invisible. There are two ways a computer can store an accented letter like "é": as a single character, or as an "e" followed by a separate accent mark. They look identical on screen. But our files were saved one way, while the addresses arriving from browsers used the other. To the computer, "sécurité" and "sécurité" were two different words.
The fix was to normalise both sides to the same form before comparing them. But the lesson was bigger than the fix: if your product only works in languages without accents, it doesn't really work in French. You only find bugs like this if you build in the language from the start, and test it like a first-class citizen.
The language switcher that changed nothing
Another one: a visitor picked a new language from the menu, and the address in the browser changed, but the text on the page didn't. The site had served a cached copy of the page in the old language.
The fix was a single line that forces the page to reload its content after a language change. Finding it took much longer than writing it. Language switching is a feature people use once and remember forever. If it fails, they assume the rest of the product is just as broken.
In the app: follow the phone, remember the choice
On the website, the address tells us which language to use. In the passenger app, we follow a simple order:
- If you've chosen a language in the app, we use that, and remember it.
- If you haven't, we use the language your phone is set to.
- If we don't support that language yet, we fall back to English.
Every part of the app (home, booking, rides, settings and onboarding) is organised the same way in every language, so a new language can be added by filling in the same structure, not by reworking screens. That's how Lingala joined French, English and Swahili in the passenger app: as a fourth set of the same pieces, not a fourth project.
We're not finished. Parts of our driver experience still speak only English, and closing that gap matters to us as much as any feature on our roadmap. Building in several languages from day one doesn't mean you never fall behind. It means catching up is a matter of filling in the gaps, not rebuilding the product.
Language is part of safety and trust
It's tempting to think of language as polish: something that makes a product feel nicer. For a mobility platform, it's much more than that.
Safety depends on it. The plate check, the driver's name and photo, the warning when a driver disconnects mid-trip: each one only protects someone who understands it.
Fairness depends on it. A fixed price only builds trust if the passenger can read it, and read what it includes, before they confirm.
Dignity depends on it. When we tell a driver why a document was rejected, as we explain in Every "No" Needs a "Why", the reason has to be in a language they read comfortably. That's one reason we use a fixed list of rejection reasons rather than free-text notes: a known reason can be translated carefully once, while a note typed in a hurry can't.
A platform that speaks to people in their own language is telling them something before any feature does: this was built for you.
Our opinion, stated plainly
If you're building a product for more than one language, or plan to someday, here's what we'd tell you:
- A safety feature you can't read isn't a safety feature. Language is part of how your product protects people, not decoration.
- Translation is an architecture decision. Make it on day one, while it's cheap. Later, it's a rewrite.
- Give every language its own address. Real URLs, real search results, real links people can share.
- Keep every language the same shape. Identical structure makes gaps visible and new languages easy to add.
- Treat every language as a first-class citizen. Accents, date formats and language switching are where you find out whether you really do.
- Write, don't just translate. Content people care about deserves to read naturally in their language.
Go back to that passenger at the roadside.
This time, the message appears in the language she reads every day: Vérifiez la plaque. Hakiki namba ya usajili. Check the plate. She reads it in a second, looks up, compares the numbers and gets on, knowing the vehicle is the right one.
Same code. Same feature. The only difference is that this time, it was built for her.

