News
The operator group says its AI Tutor is the first of many digital services to be enhanced with its local language AI models
This week, mobile network operator group Veon has announced the launch of its newest GenAI product: an AI Tutor aimed at teaching users the Kazakh language.
Powered by the Kazakh language AI model KazLLM, AI Tutor provides six gamified learning modules focussed on language learning, supported by a GenAI assistant. Additional modules covering other disciplines are planned for the future.
AI Tutor will be available to Beeline Kazakhstan customers via the company’s super-app Janymda, which currently has 4.9 million monthly users.
“AI Tutor makes learning more accessible to speakers of Kazakh and those who would like to learn Kazakh language by turning AI capabilities into a locally relevant product, seamlessly integrated into Beeline Kazakhstan’s leading digital platform Janymda,” explained Kaan Terzioglu, CEO of VEON Group in a press release. “This is exactly what we intended to offer to our customers when we positioned our AI1440 as a tool that augments human capabilities, and we are determined to put similar technologies to the service of millions of users in our frontier markets through our widespread and accessible digital services and solutions.”
Veon has been espousing the benefits of creating local language LLMs for some time now. Beeline Kazakhstan launched its software company QazCode in 2023, with the company quickly put to work co-developing KazLLM alongside the Institute of Smart Systems and Artificial Intelligence at Nazarbayev University, and Astana Hub, with the coordination of the Kazakhstan Ministry of Digital Development. KazLLM was formally launched in December 2024, with Veon saying it would use this resource to create “augmented intelligence tools” to “enhance the daily lives” of its customers in Kazakhstan.
Speaking to Total Telecom in an interview at MWC Barcelona earlier this year, Veon’s Group Chief Digital Operations Officer, Lasha Tabidze, explained the rationale for the creation of KazLLM.
“There is a big misconception around using LLMs that everyone speaks English and can prompt effectively in English. That simply isn’t true,” he explained. “There are thousands of languages in the world, but there are only seven that are fully resourced [with public online data], so these are the ones the big players are fine tuning. Less-resourced languages are not getting the attention they deserve.”
According Tabidze, KazLLM already offers answers “40% more precise” than ChatGPT in the Kazakh language, providing local users with a more nuanced experience.
At a press event this week, Tabidze re-emphasised that Veon was “not in competition with the hyperscalers” when it came to LLM creation, since these tech giants have little incentive to build AI models for relatively small markets.
“It’s a different race, a different game,” he said. “This is about supporting our customers and creating solutions for them in their language.”
“AI can be a great tool to shrink the digital divide. But if we don’t make these tools available in local languages, then we risk making that divide even bigger,” he added.
Of course, there is also a significant commercial upside to developing local language LLMs too – particularly where they face little direct competition from the hyperscalers. Veon has been greatly expanding its range of digital services in recent years, from digital healthcare platforms to ride-hailing apps, all of which can be significantly enhanced by incorporating local language AI models.
“We have 40 million plus users out of our super apps, 37 million plus of our financial services, and 28 million users in Ukraine registered on our Helsi platform,” said Tabidze. “We’re using all the power and knowledge we gathered […] to create augmented intelligence agents and bring it into the hands of people.”
Currently, Veon’s digital services revenue stands at just over 14%. Tabidze says Veon’s goal is to increase this figure to 50%, a goal he thinks is achievable in “3–5 years”.
Finally, in addition to leveraging their LLM for their own digital products, Tabidze says Veon plans to offer its LLM to enterprises on an ‘as a service’ basis.
“KazLLM is a foundational model,” he said. “If a company in Kazakhstan decides that they want to train this model themselves on their data – not just use this as a ready agent, but train it – then that’s where LLM as a service comes into play.”
Veon is currently working on additional LLMs in languages including Uzbek, Begali, Ukrainian, and Urdu. Additional LLMs for minority languages in each of Veon’s operating markets are also being created.
Keep up to date with the latest telecoms news by subscribing to our newsletter
Also in the news:
SWR deploys Europe’s first ’Rail-5G’ Wi-Fi
BT accelerates fibre rollout amid cost cuts
AT&T agrees $5.75 billion deal for Lumen’s consumer fibre assets
