For millions of Africans, access to digital services remains closely tied to the availability and cost of internet connectivity. While artificial intelligence is transforming how people communicate across languages, many translation tools still depend on an internet connection, leaving users with limited connectivity at a disadvantage.
Kenyan software developer Ruth Shallom is taking a different approach with Lugha Desk, a free, open-source Windows application designed to translate text offline, with experimental support for 19 African languages, including Kiswahili, Luganda, Kinyarwanda, Somali, Wolof and Yoruba.
Unlike cloud-dependent translation services, Lugha Desk comes with its AI translation model already bundled into the application. The approximately 1.1 GB download allows users to install the software once and continue translating without internet access, while keeping their text on their own computers.
The application also highlights the opportunities for African developers to build practical AI solutions tailored to local challenges. Rather than developing a new foundational AI model, Shallom has focused on making existing technology more accessible, addressing translation errors and simplifying installation for everyday users.
But building an alternative to established platforms such as Google Translate comes with challenges, particularly around translation accuracy, limited African-language datasets and the resources required to improve AI models. Lugha Desk is available for Windows through the Microsoft Store. The application is free and open source, with experimental translation support for 19 African languages and offline functionality after installation. Readers can access the application here: Download Lugha Desk from the Microsoft Store.
In this edition of TechMoran’s CEO Weekends, Shallom discusses the inspiration behind Lugha Desk, the technical challenges of bringing offline AI translation to Windows, competition with global technology companies and her ambitions to improve African-language translation over the next five years.
Who is Ruth Shallom, and what inspired you to develop Lugha Desk?
I’m Ruth Shallom, a developer in Kenya. A translation app can list your language and still be out of reach the moment your data runs out. I built Lugha Desk because I wanted the connection you use to install the app to be enough. After that, the translation should happen on the laptop.
That idea shaped the app. I put the model inside the Windows download, even though it made the package about 1.1 GB. Once installed, Lugha Desk can keep translating when the connection drops.
What problem does Lugha Desk solve, and how is it different from Google Translate?
Lugha Desk is built for a specific situation. You have a Windows PC and text you need to understand, but you can’t count on an internet connection.
Google Translate is a familiar point of comparison. With Lugha Desk, the translation model comes with the app, it runs on the PC, and the text you enter stays on the device. The focus is offline desktop translation involving African languages.
Why did you focus on African languages, and how did you select the 19 language options?
I wanted African-language translation to be available offline on Windows. The 19 options come from the languages the bundled TranslatePsy-AfriSLM model supports alongside English.
They include Kiswahili, Luganda, Kinyarwanda, Somali, Wolof, and Yoruba. I describe them as experimental because translation quality varies by language and sentence.
How does Lugha Desk work offline, and what were your biggest technical challenges?
The Microsoft Store package includes the translation model and the software that runs it. After installation, Lugha Desk loads the model from disk and translates using the computer’s processor. There’s no separate model download.
One of the biggest practical challenges was getting the complete package through Windows Store checks while keeping installation straightforward. Bundling the model makes the first download larger, but it also means the app has what it needs to translate after installation.
How are you addressing translation accuracy and the shortage of African-language datasets?
I start with specific mistakes and check what each fix actually improves. For example, Lugha Desk uses a small local dictionary for exact matches on selected Wolof and Luganda greetings.
Longer or different sentences still go through the model. That gives me a way to fix a known phrase without claiming that I’ve improved every translation in that language.
Lugha Desk uses an existing model. My work so far has focused on packaging, testing, and targeted fixes. The shortage of high-quality datasets remains a wider challenge, and evaluating more examples with fluent speakers would help show where the translations preserve meaning and where they still fall short.
How many users does Lugha Desk have, and how are you funding its development?
Lugha Desk translates offline after installation. A download count can tell me how many times the app was downloaded, but it doesn’t tell me how many people actively translate with it.
I’m keeping usage figures and funding details private for now.
Who are your biggest competitors, and what gives Lugha Desk an advantage?
Google Translate is the tool many readers will know. The value of Lugha Desk depends on what you need.
If you want to translate text on a Windows PC while offline and keep that text on your own device, those are the conditions the app was built around.
It’s also free and open source. I’d judge it by how well it serves that use case, including the limits of its experimental language options.
Can African developers compete with global AI companies, and what challenges are holding them back?
Yes. A developer who understands the conditions people actually work in can make choices that matter. With Lugha Desk, that meant treating an unreliable connection as a design requirement from the start.
For African-language translation, one of the hardest things is getting enough good examples and having fluent speakers evaluate the results.
A sentence can sound natural and still say the wrong thing. Knowing where that happens is part of building a tool people can use with confidence.
What is your business model, and where do you see Lugha Desk in five years?
Lugha Desk will stay free and open source, and translation will continue to work offline.
Over the next five years, my priority is improving accuracy. I want to find more cases where a translation sounds fluent but changes the meaning, fix what I can, and be clear with users about where the app still falls short.
