We are pleased to announce that our recent research paper, "Towards Reliable Multi-Agent Unit Test Generation through Verification and Iterative Refinement," has been accepted for presentation at The International Conference on Next Generation AI Systems (NGEN-AI 2026).
This work introduces MAVerIT, a novel multi-agent framework designed to address the critical reliability issues – such as compilation errors and execution failures – frequently encountered in unit tests generated by large language models (LLMs). Implemented in Python, MAVerIT utilizes a graph-based orchestration architecture to generate, validate, and iteratively refine JUnit 4 tests for Maven-based Java projects. By leveraging targeted feedback from build systems, code coverage metrics, and mutation scores, the framework ensures that only fully functional tests are retained for regression suites.
For more information about the conference, please visit https://www.ngen-ai.org/