Kenyan startup Abantu AI has developed a deep learning technology in the field of natural language processing (NLP) that translates from major world languages to indigenous African languages.
Developed by James Mwaniki, also CEO of MoVAS, a company that offers micro-lending services to telecom companies in Africa and Asia, Abantu AI launched last September, and is building AI-driven linguistic solutions for translating speech and text to and from local African languages into English or other international languages.
Currently, the deep learning model translates from most of the major world languages to Kikuyu and Kiswahili, languages spoken in Kenya and the greater East Africa region. Work is ongoing towards translation to other local African languages.
“My main motivation for this project is to build a tool that will help caregivers and health workers to better understand their patients where patients cannot speak English or Kiswahili, and in cases where the health workers themselves do not speak or understand English or Kiswahili well enough,” Mwaniki said.
NLP, he said, has developed significantly elsewhere in the world, but has been lagging behind in Africa.
“Africa is a language-rich continent with a large number of people who cannot communicate in the main languages of the world, like English and French. It is thus difficult for such people to consume and give knowledge to and from the outside world without the help of a third party,” said Mwaniki. “This is the gap we saw and we decided to build tools to address this challenge.”
The self-funded Abantu AI has developed a working proof of concept, and received “better than anticipated” market response from the media, government and health sectors.
“The next phase of our business will be to refine our products, tailor them to specific industry needs, and marketing,” Mwaniki said.
With that in mind the startup is now beginning a fundraising journey.
“We are mostly looking at grants but are open to other types of funding. So far, we received grants from Amazon which helped us in offsetting training and hosting costs for our AI models, as this is usually the most expensive part of AI industry,” said Mwaniki.
Abantu AI also plans to begin offering its services in other African countries by the end of the year. The startup monetises via a subscription model that sees clients sign up for its services on a usage or periodic basis.
“We also have clients who have asked us to custom make services for them, so this too is a different revenue model for us. Since we are still early on, we are not yet at the revenue realisation stage as much of the work has been in developing PoC and testing the market, so our clients at the moment are on trial basis,” said Mwaniki.
He said AI was still a young industry worldwide, and even more so in Africa, meaning it came with specific challenges.
“It requires specific skillsets which are hard to come by locally. It is also multi-disciplinary hence requires rich knowledge in wide field sets. Tools to build and train models are not well standardized and it takes some significant learning curve to be able to get any models trained properly. And then there’s the cost of training models which is prohibitively high for most startups,” said Mwaniki.
“All these factors contribute to raising the entry-level barrier for a lot of startups, which we are hoping to help address in future by sharing our learning processes with local institutions and organisations that would be keen to venture into this industry.”
