VEON develops local language LLMs to grow AI digital services

VEON Group is determined to augment its digital services offerings with AI that speaks the local language in each of its six markets. The operator has already developed a large language model in Kazakhstan, called KazLLM, that is powering a new educational app. Now it plans to replicate the model across the group as part of its digital services growth strategy.

“It's important that we deliver AI in the languages people speak in [our] markets. There is still a bias in the world that AI only speaks English,” said Lasha Tabidze, Chief Digital Operations Officer at VEON, speaking to journalists during a recent media briefing in London.

VEON has more than 150 million customers across Pakistan, Ukraine, Bangladesh, Kazakhstan, Uzbekistan, and Krygyzstan.

KazLLM was created by QazCode, the group’s software development company that was spun out from local operator Beeline Kazakhstan as a subsidiary in 2023. With a team of 700 employees, the company develops digital products and services for Beeline, VEON’s other operators, as well as local and international enterprises.

QazCode worked with Beeline, universities, research institutions and the Ministry of Digital Development to train KazLLM using a wealth of local data. The development was also supported by GSMA Foundry and Barcelona Supercomputing Center. The first application to use KazLLM is AI Tutor, a Kazakh language learning assistant for children and adults.

Compared to ChatGPT, KazLLM is 40% more precise for Kazakh language prompts, according to VEON.

AI Tutor is embedded in Beeline’s super app, Janymda, which has nearly 5 million monthly active users. The operator is also considering offering KazLLM as a service to enterprises that can train the model with their own data to tailor it to their needs.

KazLLM is a first move for VEON and it plans to share QazCode’s foundational model and expertise with local teams in its five other markets so that they can develop their own LLMs in Bengali, Urdu, Uzbek, Ukrainian, as well as multiple dialects.

“That’s the power of [VEON] group. We can easily replicate and share knowledge if something is successful, for example in healthcare or financial services,” said Tabidze, emphasising the importance of leveraging local talent and resources in each market, and upskilling where needed.

Local AI role for telcos

A handful of other telcos are also working on local language LLMs. For example, Orange has partnered with OpenAI and Meta to use their respective Whisper speech model and Llama text model to build customized models that understand regional languages. The operator is starting with Wolof and Pulaar, which are spoken by 16 million and six million people, respectively, and will offer the open-source AI models for free for non-commercial use, such as for public health or education services. Orange has said it aims to enable models for all languages across its African footprint of 18 countries.

The Global Telecom AI Alliance (GTAA) – comprising SK Telecom, Deutsche Telekom, e& Group, Singtel and Softbank – is also incorporating local languages into its mission to build telco-specific LLMs.

VEON’s Tabidze believes telcos are well-placed to develop and distribute local language LLMs. Given that telecoms is one of the most regulated sectors, telcos are starting from a trusted position in terms of data security and sovereignty.

“When we train our data, our environment is one of the most secure in the country…That’s why telecom operators have to step up and use this power,” he said.

By starting with basic AI models, like Gemini or Llama, telcos can build a layer on top that is created by local people and local data that is secured and trusted, he explained, as VEON did with KazLLM.

While stressing that telcos are not in competition with hyperscalers here, Tabidze said they have two clear advantages: large amounts of useful data and “the cheapest” digital distribution channel. “Telcos can put any product to the mass [market], in the hands of people, faster than anyone else,” he said.

AI-infused digital operator

VEON views AI as “augmented intelligence” that can empower its customers and support its digital inclusion mission. Its AI strategy is focused on adding value to its portfolio of digital services that spans education, entertainment, financial services, and healthcare.

The operator hopes that enhancing its services with local language models will help to ensure that no one is left behind as AI usage grows.

“If we don’t do this, especially in frontier and developing markets, the digital divide will be big. AI can narrow the digital divide or enlarge it so much that there will be no possibility to [keep up] with developed markets’ development pace. And that’s what frightens us. We want everyone to have … the power of AI,” said Tabidze.

There are three dimensions to VEON’s AI strategy: putting AI into people’s hands and in their language (e.g., AI Tutor); leveraging AI to improve the experience of its digital services; as well as creating commercial offerings for enterprises from its internal use of AI. For example, if the telco has developed agentic AI tools for its own legal or human resources tasks, these can be packaged into services offerings for third parties.

VEON launched its digital operator transformation strategy four years ago, called DO1440, followed by a complementary AI strategy, dubbed AI1440. The 1,440 figure is the number of minutes in a day, as the operator’s aim is to be a digital operator that engages customers every minute of the day.

At the end of March 2025, VEON had 125 million monthly active users on its digital services and platforms, of which 32.4 million are digital-only customers who do not subscribe to telecom services.

Digital services accounted for more than 14% of VEON’s total revenue in the first three months of this year, and the operator’s target is to increase the share of direct digital revenue to 50% in the next three to five years.

Examples of VEON’s digital portfolio include JazzCash in Pakistan and healthcare app Helsi in Ukraine. The operator has also launched a supper app in each market, which have 45 million monthly active users in total.

When asked how VEON has changed since the launch of the digital operator strategy, Tabidze said the most important shift has been cultural, especially in the way products are created. That is, the operator has moved away from a “telco mindset”, which centrally develops products and pushes them down to the markets.

“We truly believe that ‘one size fits all’ is not working. We always look into the communities… What problem can we solve? What benefits can we bring? And then, create products specifically tailored to the markets we operate in,” he said.