Synthetic Trace Generation for Intelligent Modeling Assistants

This technical report presents an extension of the MASTER-LLM framework, which integrates synthetic trace generation of modeling operations to enhance intelligent modeling assistants (IMAs) in the context of Model-Driven Engineering (MDE). The document outlines the interplay between generative AI models and MDE practices, highlighting the challenges posed by the limited availability of training data due to privacy and security concerns. The proposed framework combines various MDE tools to support both industrial and academic practitioners. The report details the methodology employed, which includes the use of different large language models (LLMs) to generate synthetic datasets of modeling events. Experimental results are provided from two modeling environments, CAEX and HEPSYCODE, demonstrating the effectiveness of synthetic traces in less complex domains while emphasizing the necessity of human-based operations in more intricate scenarios. The findings suggest that a mixed approach may be beneficial when training data is scarce, and the report concludes with a discussion on the implications of using generative AI in industrial modeling contexts.