Researchers at the University of Klagenfurt have developed 'CheckMate', an AI-supported technology that can independently generate algorithms for complex combinatorics and optimization problems. The solution should deliver better results for industrial applications than previous methods. A patent has already been filed for the technology.

Such systems could be used, for example, in the planning of industrial plants, in logistics or in railroad safety systems. The aim is to find the best possible variant from a very large number of possible solutions.

The starting point of the research is so-called "hard problems", where conventional methods quickly reach their limits. As an example, the team cites greatly enlarged Sudoku variants with hundreds of thousands of rows, columns and number combinations. According to the researchers, the computing and memory requirements of such tasks significantly exceed the capabilities of conventional approaches.

Gerhard Friedrich, Professor at the Institute for Artificial Intelligence and Cybersecurity at the University of Klagenfurt, and his team have been working on such problems for years. The research team consists of Veronika Semmelrock, Benedetta Strizzolo, Francesco Zuccato, Patrick Rodler, Konstantin Schekotihin and Gerhard Friedrich.

With CheckMate, the researchers are now presenting a system for automated evolutionary code generation. The technology is to be presented at the International Joint Conference on Artificial Intelligence 2026.

"In recent years, we have had to realize that we humans are not able to develop tools for such difficult problems that come closer to the vision of an AI that actually solves hard problems independently," explains Gerhard Friedrich.

The system can be integrated into the open source framework OpenEvolve. With the help of logic checks, efficiency tests and continuous feedback, the AI gradually develops its own solution algorithms. According to the researchers, CheckMate only requires a description of the rules and a few example instances.

"Our team has now succeeded in demonstrating that the programs generated with CheckMate outperform the current leading technologies and achieve better solutions for hard problems with direct industrial relevance than was previously possible. In particular, CheckMate has also made it possible to solve particularly hard, previously unsolved problems for the first time. We therefore seem to have entered the age in which AI is at least on a par with humans when it comes to solving complex problems," says Gerhard Friedrich, summarizing the results: "LLMs + Logic + Evolution = Superhuman Programming."