Text-to-BPMN 2.0 – From Natural Language to Executable Processes

Business process modeling through BPMN 2.0 is essential for formalizing organizational workflows but remains a technically demanding activity. This thesis introduces Text-to-BPMN2.0, a system that simplifies this task by leveraging large language models (LLMs) to automatically transform natural language descriptions into formal BPMN diagrams. Users can input processes via text, files, or images and select between multiple model configurations to influence generation. The backend orchestrates model interactions, while the frontend renders diagrams with integrated validation and editing capabilities. Evaluation tests show that model selection significantly impacts output variability and consistency, underscoring the importance of semantic validation and user-driven refinement. The system demonstrates the feasibility of combining generative AI with formal modeling frameworks, offering a practical and extensible solution that lowers the barrier to BPMN modeling, while maintaining flexibility and control. The results highlight promising directions for making process modeling more accessible and adaptive to diverse business contexts.