Enhancing declarative business process management availability through generative AI

Business process models serve as crucial artifacts in modern organizations, enabling stakeholders to understand, analyze, and improve their operational workflows. While traditional imperative process models are well-represented in public repositories, declarative models—particularly those capturing multiple process perspectives—remain scarce due to privacy concerns and the proprietary nature of business operations. This scarcity significantly hampers research, education, and innovation in declarative process management. We address this challenge by introducing Terpsichora, a framework that leverages large language models to generate synthetic multi-perspective declarative (MP-Declare) process models. Our strategy combines input formulation and constrained generation techniques with automated validation mechanisms to ensure the generation of diverse, realistic process models while maintaining structural validity and semantic coherence. Through comprehensive evaluation of 2000 synthetic models across two generation approaches, we demonstrate that our framework successfully generates models with diverse complexity metrics and high labeling convention compliance. The generated models effectively support various analytical tasks, including process mining and conformance checking, while preserving essential business domain characteristics. This research establishes a novel pathway for addressing the availability challenge in declarative process modeling, enabling researchers and practitioners to access diverse, high-quality process models.