Ukraine is creating its own AI: how this will change government services

This was reported by RBC-Ukraine with reference to Danilo Tsvok, head of the AI department at the Ministry of Digital Transformation.

Ukraine is preparing for a massive technological leap that could change not only public administration but also the very approach to citizen-government interaction.

One of the new priorities of digitalisation is the creation of a national language model and the introduction of intelligent agents. These steps are intended to strengthen data protection and ensure the development of proprietary technologies that take into account the Ukrainian context.

The Ministry of Digital Transformation has set a goal to become one of the world's top three leaders in the development and implementation of artificial intelligence in the public sector by 2030.

This course involves a transition from a digital state to a so-called agentive state. This is a model in which public services will become proactive.

According to Danylo Tsivok, instead of people searching for the right service themselves, special AI agents will predict citizens' needs and automatically offer solutions.

An example of this is the birth of a child. In this case, the system will be able to suggest how to complete the paperwork and offer the necessary social options, such as the "Baby Package".

One of the key projects in this area is the creation of a large language model, often referred to as Ukrainian ChatGPT.

At the same time, unlike foreign products such as Gemini or OpenAI's developments, the Ukrainian model is being created with deep consideration of the local context, history, and culture.

The article emphasises that this is particularly important in the context of information warfare, as such a model must clearly understand the political realities, the status of Crimea and the reasons for the war, and not be susceptible to manipulation by the enemy.

In addition, according to the Ministry of Digital Transformation, Ukraine's own LLM is two and a half, and in some cases three times cheaper than foreign counterparts, making it more cost-effective for the state and business.

A separate emphasis in the development is placed on data security. For this purpose, a proprietary infrastructure known as AI Factory is being created.

The article notes that the use of popular global chatbots usually involves the transfer of information to foreign cloud servers, which is unacceptable for working with sensitive data in the fields of defence, medicine and the public sector.

The Ukrainian language model should enable information to be processed within the country so that personal data does not leave the national perimeter.

The Diya.AI system already uses masking technology that anonymises user data, and communication histories are encrypted with unique keys.

The introduction of artificial intelligence has already yielded initial results in reducing bureaucracy.

The Diya support service independently processes more than 90% of requests, providing answers within five seconds.

Algorithms also help harmonise Ukrainian legislation with European Union standards by analysing over 30,000 legal acts. The article notes that such work would take decades to complete manually.

Legal examination of documents, which previously took weeks, has now been reduced to 72 hours thanks to artificial intelligence.

Despite fears of mass layoffs, the Ministry of Digital Transformation emphasises that artificial intelligence does not replace people, but changes their role.

Specialists who previously performed routine operations are being retrained and becoming tutors or managers for training neural networks.

The ministry explains that the number of public services is constantly growing, and technology makes it possible to scale processes without endless staff expansion.

Currently, the development of Ukrainian LLM is in an active phase in partnership with Kyivstar.

Critical components of the architecture, including a tokeniser, have already been created. The collection of large data sets from scientific institutions, universities, and the media is also ongoing.

The presentation of the first version of the model is scheduled for the end of spring 2026.

After testing is complete, it is planned to make it available for free use in open-source format. This should enable scientists and businesses to create their own specialised products based on it.