LLM-based System Design Automation (LSDA) Using Generative AI: Exploring a Formalized Framework Based on Experiments

Since the 1968 NATO Conference on Software Engineering laid the groundwork for systematic development of software systems, much effort has been dedicated to automating software design including Computer-Aided Software Engineering, Model-Driven Development, low-code/no-code platforms, and AI-driven tools like ChatGPT from OpenAI. Today, Generative AI offers an unprecedented opportunity for converting user requirements in natural language into system specifications via a Large Language Model (LLM). In this paper, we validate LLM-based system design automation (LSDA) for modeling ontology, workflow, entity-relationship diagrams and other artifacts on real-world examples. We found that ChatGPT is able to produce these artifacts with surprisingly high accuracy and quality even with a zero-shot prompting approach and simple prompts with little prompt optimization.