Answers are grounded in this paper and cite their sources.
Context: The increasing complexity of modern software systems poses challenges for traditional development methods. Manual conversion of specifications into source code often leads to inefficiencies, inconsistencies, and higher maintenance costs. Model-Driven Engineering (MDE) aims to address these issues by elevating the level of abstraction. However, its adoption is still hindered by methodological challenges, tooling limitations, and the complexity associated with Domain-Specific Modelling (DSM). Objectives: This study introduces Simplified Domain-Specific Modelling (SDSM), a functionally comparable but more accessible counterpart to DSM. SDSM operates without domain-specific languages, relying instead on simple data structures and a clearly defined implementation procedure to reduce implementation complexity and improve reusability in application development, rendering it particularly advantageous for the systematic generation of repetitive software artefacts. Methods: We integrate Large Language Models (LLMs) as formal code generators within the SDSM framework. This integration transforms LLMs from ad-hoc programming assistants into structured, repeatable components of the software engineering process. To the best of our knowledge, this proposal is the first to simplify DSM by eliminating a Domain-Specific Language (DSL) and utilising LLMs as code generators in this process. Results: Two real-world experiments compared LLM-based generators to traditional Template-Based Code Generation and programmatic approaches. The evaluation concerned code quality and productivity. Results indicate that, in specific contexts, LLMs can effectively replace conventional generators, achieving competitive code quality and productivity improvements while reducing repetitive, non-creative coding tasks. Conclusion: Our findings provide empirical evidence for the underexplored intersection of MDE and AI, highlighting both the potential and current limitations of LLM-driven model-based software engineering. We provide practical demonstrations that the SDSM method and LLMs form an integrated software development framework that can be utilised in real-world software.
No takes yet. Share an insight, caveat, or question.
Bochenek et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: