Automated Machine Learning (AutoML) is a rapidly evolving research area, gaining significant popularity among diverse user groups with varying requirements. A fully automated process is very user-friendly to non-expert users, with the disadvantage of missing transparency about the procedure and decision-making. Transparency about the components used, the best results, and a comparative analysis of the performance improvements of the different AutoML configurations are missing. Furthermore, the enclosed architecture of the most open-source AutoML frameworks prevents researchers from a simple modification or implementation of new, domain-specific ML components. To address these challenges, this study presents an early prototype of an Evolutionary Algorithm (EA) based AutoML design pattern. This design pattern includes a log-file concept for transparency of the decision process and an overview of performance improvements. The architectural pattern enables the different structural components to be extended intuitively. The first experiments were performed using ten datasets that differed in terms of the number of features and data types. The results show that maximizing preprocessing improves accuracy through targeted hyperparameter optimization, demonstrating the complementary effects of the two optimization stages.