Software Modeling Assistance with Large Language Models

Software modeling requires a challenging combination of expertise in both domain knowledge and modeling formalisms. Existing methods often fail to provide effective, general modeling assistance. This research introduces a novel approach using large language models (LLMs) to enhance software modeling. Utilizing few-shot prompt learning, our method supports various modeling activities without extensive training data. Initially focusing on static and behavioral formalisms like UML diagrams, we aim to extend this to other paradigms and integrate it into the Model-Driven Engineering (MDE) pipeline. Additionally, we aim to assess productivity, model quality, and accuracy when receiving real-time, context-aware suggestions during modeling tasks.