Matěj Klouček’s master’s thesis, entitled Evaluation of Code Generated by LLM-Based Coding Agents, was supervised by Ing. Jan Šedivý, CSc. It addresses one of the current challenges in using artificial intelligence for software development: how to ensure that AI-generated code not only looks correct but actually works.
Today’s AI agents can independently create and modify entire web applications. However, the resulting software may contain errors that only become apparent when the application is used in practice. Matěj Klouček therefore investigated whether coding agents can test their own work, identify shortcomings and subsequently correct them.
His research compared three approaches: conventional code generation without ongoing verification, automated testing performed by the coding agent itself, and the use of a separate evaluation agent that checks the application in much the same way as a real user. The experiments showed that incorporating automated verification can significantly improve the functionality of the resulting software, particularly when more powerful AI models are used. A key finding is that weaker models do not necessarily fall behind because they are unable to fix errors, but because they cannot reliably detect them on their own. When a more powerful model evaluates their work, their success rate more than doubles. This opens the door to more cost-effective solutions in which a less expensive model writes the code while a more powerful model handles verification.
The findings could contribute to the development of more reliable AI coding tools and reduce the number of errors in automatically generated applications. The research was conducted as part of a team at the Czech Institute of Informatics, Robotics and Cybernetics (CIIRC CTU), which is developing coding agents for the Amazon Nova AI Challenge 2026.
Seven Years, Seven Medals: FEE CTU Aims to Continue Its Remarkable Winning Streak
Matěj Klouček’s place in the IT SPY 2026 finals confirms the exceptionally strong position of FEE CTU graduates in the competition. Since 2019, they have reached the podium in every edition, securing five first prizes and two second prizes over seven years.
Last year’s winner was Ing. Aneta Furmanová, a graduate of the Medical Electronics and Bioinformatics programme, whose master’s thesis focused on using machine learning to model the properties of sonic crystals. Her research could contribute to the development of more effective technologies for reducing noise pollution.
In 2024, Jiří Janota took second place with a project focused on robotic mapping of honeycombs. A year earlier, Pavlína Koutecká achieved the same result with a thesis on optimising surgical procedure scheduling. First prizes in previous years went to Jiří Ulrich (2022), Jindřiška Deckerová (2021), Jan Bayer (2020) and Šimon Mandlík (2019).
FEE CTU is thus among the most successful faculties in the history of the competition. We will be keeping our fingers crossed for Matěj Klouček as he seeks to follow in his predecessors’ footsteps and extend this remarkable record.
Support Matěj Klouček in the Public Choice Award!
IT SPY (IT Student Project of the Year) is a competition recognising the best master’s theses in computer science and information technology, with universities from the Czech Republic and Slovakia participating each year. An expert jury selects eight finalists who compete for the main prize. Contestants also have the opportunity to win the Public Choice Award.
Voting for the Public Choice Award is open until 23 November 2026 at www.itspy.cz. Support our graduate Matěj Klouček with your vote!
The award ceremony for this year’s competition, now in its 17th edition, will take place on 25 November 2026.