Automated unit test creation based on Artificial Intelligence

This thesis explores the application of Generative Artificial Intelligence, specifically Large Language Models (LLMs), for the automated generation of C++ unit tests, addressing the common issue of projects being developed without sufficient testing. The study focuses on overcoming LLM challenges, such as context limitation and low reliability, by developing an Adaptive Focal Context procedure for C++, to select information to use in LLM queries for generating unit tests, and implementing a check-repair mechanism to fix incorrect unit tests. The created system is validated using various LLMs, obtaining results comparable to previous studies for other languages. Despite overall positive results, challenges related to handling certain C++ functionalities and packages emerged, suggesting areas for future research. The thesis concludes with suggestions for system improvements and strategies to address identified issues.