Original article Total Telecom:Read More
News
The LLM will be carefully trained on Ukrainian data in an attempt to avoid undue Russian influence
Kyivstar and Ukraine’s Ministry of Digital Transformation have chosen Google’s Gemma model and Google Cloud’s Vertex AI to form the technical backbone of a national large language model (LLM) intended to capture the breadth of Ukrainian dialects, terminology, and history.
The project, announced today, will be operationally led by Kyivstar, the country’s largest mobile operator and a unit of telecoms group VEON.
Kyivstar and the WINWIN AI Centre of Excellence at the Digital Ministry said the model will be trained on curated Ukrainian datasets and will keep sensitive national data stored and processed within Ukraine, an explicit priority for future use in government, healthcare, and financial services.
“We are building the Ukrainian LLM on a ready-made open-source model. The main task in development is to train it on our unique data further. When choosing a model, we focused on how well it already handles Ukrainian-language texts and how controllable it is during additional training. This will help minimize linguistic and ethical risks in our LLM,” said Danylo Tsvok, Chief AI Officer at the Ministry of Digital Transformation and CEO of the WINWIN AI Center of Excellence.
Google’s Gemma was selected after an “extensive evaluation”, the partners said. Google Cloud’s Vertex AI will provide the computing infrastructure for large-scale training. Krzysztof Kaziów, Director Customer Engineering CEE at Google Cloud, commented: “We are honored that the Ministry of Digital Transformation and Kyivstar have selected Gemma as the foundation for the Ukrainian national LLM. This choice underscores Gemma’s strategic value, offering an optimal balance between performance and resources alongside its strong multilingual support. Leveraging its proven success as the base for leading Ukrainian LLMs, we are committed to supporting this vital initiative to enhance digital experience in Ukraine.”
VEON described the project as part of a broader strategy to develop local-language AI across its markets, pointing to prior initiatives such as KazLLM in Kazakhstan and an Urdu LLM in Pakistan. “Kyivstar and the Ukrainian Digital Ministry have taken a major step forward today. With a sovereign Ukrainian LLM, Ukrainian consumers, businesses and government institutions will be empowered to integrate cutting-edge technologies using augmented intelligence that truly speaks Ukrainian and understands Ukraine,” said Kaan Terzioglu, CEO of VEON Group. “We have a responsibility to bring the benefits of augmented intelligence to the countries we serve , through large language models trained not only on words, but on local context.”
Technical work will include optimising Gemma for Ukrainian, refining its tokenizer, and creating benchmarks for fine-tuning and application-specific adaptation. Intended use cases span regulatory and legal analysis, education, finance and healthcare.
While not explicitly stated by the partners, project’s cultural significance for Ukraine should not be understated. Much of the data used to train existing Ukrainian LLMs was produced under substantial Russian influence, leading to significant bias. A domestically-led LLM could correct this influence, creating a model that better represents Ukrainian linguistic, historical, and civic perspectives.
Keep up to date with all the latest telecoms news with the Total Telecom newsletter
Also in the news
Connected Britain Award winners 2025 announced!
Netomnia announces ‘powerful and ambitious’ rebrand ahead of Connected Britain
VodafoneThree drops Samsung, relies on Nokia and Ericsson for £2bn network upgrade