LLM-Based Recommender Systems for Violation Resolutions in Continuous Architectural Conformance

Software architectures are fundamental to the development, evolution, and quality of software-intensive systems. Architectures rarely exist in isolation, but instead adhere to overarching structures such as architectural patterns and styles, frameworks, software product line architectures, and reference architectures. To fully leverage the benefits of these structures, conformance between them is essential, enhancing interoperability, reducing costs through reusability, mitigating project risks, and facilitating the adoption of best practices. In our previous work, we introduced the concept of continuous conformance and focused on detecting architectural violations using a model-driven engineering approach.In this paper, we extend our previous work by proposing a large language model-based recommender system into the model-driven tool to suggest resolutions for architectural violations. Leveraging large language models, we reduce the accidental complexity of model-driven techniques by combining the reasoning capabilities of large language models with the formalization of architectures as (meta)models. We evaluate the success rate using two large language models and architectures from the IoT domain, including one reference architecture and four software architectures that we manually mutate in 16 faulty architectures. The results demonstrate the system's effectiveness in providing intelligent, context-aware recommendations for restoring architectural conformance.