In recent years, large language models have delivered a sharp leap in working with text and data. Using an example from public administration, IBM showed that the decisive factor is not just “which model,” but above all how AI is packaged into a useful application with clear quality control. The result is a tool that speeds access to verified information and improves services for farmers.
What AI can already do: from Slovak to scientific texts
In the demonstrations, the system answered in Slovak without difficulty and compiled an overview of the key events of the Slovak National Uprising. It also handled a practical math problem: it formalized the task, created a system of equations, clearly showed the steps, and checked the result. Finally, it processed a scientific article about how some pesticides make soil paradoxically attractive to bumblebees, even though it may be harmful to them; it summarized the experiment’s methodology and the main conclusions. The point was not the “wow effect,” but that AI can reliably read, summarize, and explain—exactly what organizations need every day.
How AI helps SZIF: verified answers for frontline staff
Státní zemědělský intervenční fond (SZIF) distributes subsidies, and its staff explain rules, procedures, and updates to farmers every day, for example satellite monitoring of land. The challenge is the volume of documents, laws, and training sessions, which are hard to keep track of. The solution was an internal knowledge system: the organization uploaded materials into it, and AI automatically prepared draft questions and answers from them, which were then reviewed by a subject-matter expert. The foundation is therefore a base of verified answers that can be quickly accessed via natural questions.
If the system finds a relevant, already verified answer (for example, what the “traffic light” colors in the portal mean or when additional evidence from the field needs to be provided), it displays it as authoritative. When an answer is missing, the AI proposes one—but clearly marks it as unauthorized; it is then sent to an SZIF expert for review, who either approves it or edits it. In practice, this shortens the path to information: instead of “I’ll ask my colleague,” an employee enters a question and immediately receives either a verified answer or a high-quality draft that is usually confirmed within an hour. The system thus continuously learns from real queries and raises the quality of advisory services; today it operates internally, with later stages planning to open it to farmers as well.
