ARCHITECTURE OF AN AUTOMATED TEST GENERATION SYSTEM

Abstract

The growing complexity of contemporary software systems necessitates more advanced approaches to ensuring their reliability and quality. Although unit testing remains a fundamental practice in software engineering, the development of effective test cases is still resource-intensive, while traditional metrics such as code coverage do not adequately reflect the fault-detection capability of test suites.

Recent advancements in large language models (LLMs) have enabled the automated generation of test cases; however, the consistency and effectiveness of such tests remain limited without additional validation mechanisms. In this context, the present study introduces an architecture for an automated unit test generation system that integrates generative artificial intelligence with mutation testing techniques.

Unlike conventional approaches, mutation testing is employed not only as an evaluation metric but also as a structured feedback mechanism within an iterative test generation cycle. Mutation analysis results, including information about surviving mutants, mutation operators, and code locations, are utilized to enrich LLM prompts and guide the refinement of test cases.

The proposed architecture incorporates modules for test generation, mutation testing integration, report normalization, and iterative improvement of the test suite. Special attention is given to the standardization of mutation testing reports generated by heterogeneous frameworks, enabling their consistent processing and reuse within the system.

The developed approach has the potential to enhance the effectiveness of automated testing processes and demonstrates applicability across diverse software environments and technological stacks.