Bridging the digital divide with EjoLabs

Protogene Hahirwabayo

Protogene Hahirwabayo built EjoChat, the first LLM chat in Kinyarwanda, so that people without English can use AI too. A conversation about low-resource languages and closing the AI gap.

Protogene Hahirwabayo founded EjoLabs to provide the first LLM (large language model) chat application in Kinyarwanda, the official language of Rwanda. Protogene holds a Bachelor’s degree in Electronics and Communication Engineering and a Master’s degree in Data Engineering, and has broad experience in AI across many companies, from use-case identification to agentic AI implementation.

Born in Rwanda and educated in Europe, Protogene sees daily how quickly AI can be applied to diverse scenarios. He also sees the significant gap in AI accessibility for countries with low-resource languages, where few people speak foreign languages fluently. More at ejolabs.com.

You founded EjoLabs, providing the first LLM chat in Kinyarwanda, the official language of Rwanda. Tell us more about this language, who speaks it?

Kinyarwanda is the official language of Rwanda, spoken by approximately 14 million people within the country. It is also spoken in neighboring countries such as the Democratic Republic of the Congo and Burundi. We estimate that there are around 30 million native Kinyarwanda speakers across Africa.

Historically, it is primarily an oral language; digital text production has not been a primary focus. Consequently, like many African languages, it is a “low-resource” digital language. There are very few webpages, books or videos available to train LLMs. Our language also often lacks sophisticated technical terms for subjects like mathematics, biology or physics, as these are typically taught in English in higher education. And many Rwandans do not speak foreign languages like English or French; often, only those from privileged backgrounds have the opportunity to learn them.

What motivated you to found EjoLabs and develop EjoChat, the AI chat in Kinyarwanda?

Living and working with AI in Europe shows me every day the immense opportunities that LLMs and chat interfaces provide. But back in Rwanda, people without English proficiency struggle even to register for these applications. My main motivation is to give them the same ability to interact with AI that we enjoy in Europe. This opens up a whole new world for them, whether in learning, teaching, staying informed or professional development.

How did you start with the implementation, and which challenges did you have to overcome?

Initially, we thought we would need to train our own Kinyarwanda model from scratch to ensure accurate translations. However, when we searched for digital resources in Kinyarwanda, we found almost nothing. We even contacted teachers and authors to write content for us, but it still was not enough to train a proprietary model.

As a result, we currently use Kinyarwanda-to-English translation layers. While this is roughly 60 percent accurate, it often lacks fine-grained detail and nuance. We built EjoChat around these available resources, and with 1,000 daily users we receive a wealth of feedback from native speakers to help us improve. Another challenge is financial: since we currently fund the token usage for our users out of our own pockets, we occasionally have to limit the length of responses, which can be frustrating.

What are your future plans for EjoLabs and EjoChat?

EjoChat is just the first step. Through EjoLabs, we want to build further AI tools for Kinyarwanda speakers. We see massive potential in education, to help teachers and students understand lessons in their native tongue. Another vital area is healthcare. For many Rwandans, the nearest doctor is hours away; a medical assistant on a smartphone could improve health outcomes enormously. Finally, we are looking at the business sector, where AI can help local companies gain the same efficiencies we see in Europe.

What are the current milestones for your business, and what comes next?

We currently have about 1,000 daily users, but our target is to reach 5 million in the coming years. We are spreading the word within Rwanda and are in close contact with the Rwandan Embassy in Germany. We have also presented our work to NGOs in the agricultural sector, as we see a great opportunity to lower communication barriers between NGOs and farmers.

There is a lot of enthusiasm when we present our work, because we are not just talking, we are doing. EjoChat is available right now for everyone. That ability to deliver a working product is what makes us different and makes us proud. Receiving thankful feedback from our users is incredibly rewarding.

Murakoze! Thank you, Protogene, for this great interview. We wish you the best of luck in bringing these vital AI tools to Rwanda.

AI for your use case?

Whether language, process or analysis, we bring AI to where it makes a difference in your company.

Book an introductory call or send us a message

Frequently asked questions

What is a low-resource language, and why is it hard for AI?

A low-resource language has little digital text to train on: few web pages, books or videos, and often no established technical vocabulary. Kinyarwanda is historically an oral language, so there is very little material to teach a large language model.

How does EjoChat work without a dedicated Kinyarwanda model?

EjoChat currently uses Kinyarwanda-to-English translation layers rather than a model trained from scratch. That is roughly 60 percent accurate today, and feedback from about 1,000 daily native speakers is used to improve it.

Read more