The Ministry of Youth and Sports of Ukraine has launched a service based on LLM on the Diya.Engine platform, which «speaks» Ukrainian and will assist state bodies.
The Ministry of Youth and Sports of Ukraine has launched a service based on LLM on the Diya.Engine platform, which «speaks» Ukrainian and will assist state bodies.
The automated application module using artificial intelligence, introduced by the ministry, will now automatically check applications for international youth exchange project competitions and provide advice to contestants on how to complete applications. The solution has been integrated with the «International Youth Exchange Project Competition» service, built on the Diya.Engine platform.
Previously, according to the developers, applicants often made mistakes because they had to fill in a large number of fields, the complexity of the order and the limited time. This is why they received low scores and could not win the competition. The law does not provide for the revision of applications, so the applicant does not have the opportunity to correct errors after submission. You can only withdraw the application and submit it again, and even then, only if the applicant knows that he has errors.
The AI assistant breaks the application text into logical blocks (goal, objectives, description of activities, expected results, budget, etc.) and compares their content with the criteria recorded in the competition documentation.
Now the AI assistant will identify typical mistakes made by users and provide recommendations for their elimination.
When creating it, the developers used the retrieval-augmented generation (RAG) approach, i.e. searching for relevant information from the internal knowledge base and building recommendations using a language model (LLM) taking into account the found content. This helps minimize «hallucinations» and increase the accuracy of answers. All calculations are performed on the ML Cloud infrastructure from De Novo, the data does not leave the borders of Ukraine and is not transferred to third-party services.
Oleksandr Akulenko, head of the artificial intelligence department of MK-Consulting, reported that to develop the model, they stopped at Gemma2-27b from Google, which, among the ready-made models that can be deployed in local infrastructure, showed the best result for processing the Ukrainian language (taking into account the amount of resources required, in particular video cards). «We are now experimenting with the new version of Gemma3. In general, the solution was specially developed so that it was not tied to a specific model and was more flexible. At the same time, we consider work on a national language model to be a strategically important task. Which, among other things, can be used to implement similar services in both the public sector and business. For Ukrainian business and the public sector, such solutions have a strategic context,» he added.
To deploy LLM gemma2-27b, a resource pool with NVIDIA GPU instances was used; in this configuration, the Nvidia H100 and A100 accelerators, according to the developers, are ideal for machine learning and high-performance computing.
The total setup cycle (including research, building validation logic, and integration with Action.Engine) took 3.5 months.
The service was developed by the Ministry of Youth and Sports of Ukraine within the framework of the EGAP Program, implemented by the Eastern Europe Foundation with the support of the Swiss government. Technical expertise was provided by MK-Consulting and AI/ML solutions provider De Novo
The Minister of Youth and Sports of Ukraine, Matviy Bidny, is confident that technology will help youth organizations improve the quality of their applications, reduce the number of technical errors, and increase the chances of implementing important initiatives.
