| Abstract [eng] |
Modern software systems increasingly rely on REST APIs, making effective API testing an important part of software quality assurance. However, creating executable system tests from natural language requirements is typically a manual and time consuming process. Existing approaches often generate abstract test descriptions or require additional manual implementation. This thesis proposes an approach for automatic generation of executable REST API system tests from natural language requirements using large language models and OpenAPI specifications. A prototype system was developed using Python, PydanticAI, and the OpenAI GPT-5.2 model. The proposed approach combines requirement analysis, OpenAPI based context generation, automated validation, and runtime execution into a unified workflow. The approach was evaluated using the Fixer API and the PayPal Checkout Orders API. Experimental results showed that the system successfully generated executable pytest tests, achieved full requirement and endpoint reachability coverage, and demonstrated stable execution behaviour. In addition, semantic correctness evaluation showed that the generated tests were generally well aligned with the original requirements. The results demonstrate that large language models can effectively support automatic generation of executable REST API system tests while reducing repetitive manual testing effort. |
