Trends & neue Technologien (KI o.a.)Rise of the AI Testers: generate unit-tests with AI agents
Abstract
Thu 15:35 - 16:20 Uhr | 2025
How can you create high quality unit tests for your Java code completely automatically? In our talk, we will show the inner workings of how we achieve this goal using AI Agents powered by Large Language Models (LLMs). Our tooling delivers high code coverage and extensive business logic testing across diverse Java projects.
We employ a divide-and-conquer strategy and supply the model with necessary context from the repository to understand the codebase. Additionally, a feedback loop based on containerized test execution allows the system to iteratively improve the generated tests. Here, large complex codebases present particular challenges. We will demonstrate strategies to tackle these for test generation. Finally, we will compare the results obtained with several commercial and open-source LLMs.
Dr. Marie Bieth earned her PhD in Computer Science from the TU Munich, focusing on the application of machine learning to medical imaging. She then joined TNG Technology Consulting, where she has worked in various roles, including AI Engineer, Software Engineer, and Product Owner. Marie has contributed to AI solutions for industrial, efficiency, and process-oriented applications. Her main expertise lies in Computer Vision, Generative AI and backend development. She always strives to build the right thing while also building it well.
Dr. Michael Oberparleiter is an IT consultant at TNG Technology Consulting with a background in theoretical physics. He worked mostly as a full-stack software engineer before starting a journey into various AI applications in the last year. While he covered different aspects of Generative AI including image generation in industrial applications, so far he found the field of code generation the most interesting. Some days, he wonders how this topic will affect the old Fortran codebases he worked with as a scientist.