The rise of generative AI has increased interest in automating Model-Driven Engineering (MDE) by leveraging Large Language Models (LLMs) for various tasks. However, LLMs can sometimes produce errors such as hallucinations or formatting issues, which may compromise the quality of the modeling outputs. Although recording human modeling activities can help train effective intelligent modeling assistants (IMAs), this approach often demands large amounts of data that are difficult to obtain due to privacy and security concerns. To address these challenges, our project introduces MASTER-LLM, a conceptual framework to support Modeling Assistants using Synthetic Trace gEneration of modeling opeRations through LLM. In this framework, a dedicated component within an active modeling environment records relevant operations, enabling IMAs to help designers’ complete models. Moreover, different LLMs are employed to generate synthetic datasets of modeling events, thereby accelerating the data collection process. The feasibility of MASTER-LLM is demonstrated through experiments in several domains, including electronic design automation with HEPSYCODE, automotive engineering with CAEX/AutomationML, and business process management with BPMN Designer. These experiments, based on industrial use cases from European projects and state-of-the-art benchmarks, help us evaluate how well LLMs can simulate realistic modeling operations across different fields.