Leveraging LLMs to Discover Causal Dependencies: A Case Study on a University Program

Defining a process model is essential for understanding organizational workflows and guiding future activities. While process discovery techniques can derive models from event logs, designing a process model from scratch remains challenging, as it relies on human experts to interpret requirements. This work explores process model elicitation in the educational domain, where courses in a university program are treated as activities with prerequisite relationships. We propose a methodology leveraging Large Language Models to automatically infer causal dependencies among courses by analyzing their syllabi. Using pairwise prompting with step-back and chain-of-thought reasoning, the extracted dependencies are used to build a Directly-Follows Graph, subsequently evaluated by domain experts. Experimental results show that Large Language Models can effectively identify and justify course dependencies, providing valuable insights that can enhance curriculum design and provide students with clearer guidance on course sequencing. The approach also highlights gaps in prerequisite structures, suggesting a broader application in educational planning and dynamic learning environments.