Unified Architecture Metamodel of Information Systems Developed by Generative AI
Summary: arXiv:2604.00171v1 Announce Type: cross
The rapid development of Artificial Intelligence (AI) and Large Language Models (LLMs) has significantly transformed software development life cycles (SDLC). A considerable portion of code, technical, and business documentation is now being generated automatically. However, the absence of a unified architectural framework has led to fragmented system representations across various layers of information systems.
This study addresses the challenge of establishing a unified architecture tailored for LLM-oriented applications. It is based on selected architectural frameworks proposed by subject matter experts (SMEs). The research introduces a framework structure designed to encompass essential types of architectural diagrams while supporting a comprehensive cycle of transformations, notably the “Code to Documentation to Code” paradigm.
Key Components of the Proposed Framework
The proposed unified architecture metamodel is structured into three primary architectural layers, each containing specific abstraction layers:
- High-layer (Business and Domain Understanding): This layer focuses on aligning business goals with domain knowledge, ensuring that the architecture reflects the needs and objectives of stakeholders.
- Middle-layer (System Architecture): This layer serves as the backbone of the architecture, detailing the system’s structural design and components, facilitating seamless integration and communication between different parts.
- Low-layer (Developer-layer Architecture): Targeted at developers, this layer provides the technical specifications and design patterns necessary for implementation, ensuring that the architecture is both practical and effective.
Each layer incorporates abstraction levels that enhance flexibility, making it easier to align with design principles and architectural patterns. This structured approach aims to bridge the gap between human understanding and model representation, thereby improving the overall quality of generated documentation and code.
Research Findings and Implications
Experiments conducted as part of this study demonstrated that utilizing a structured architectural context, illustrated through architectural diagrams, led to stable quality in both documentation and code generation. The findings suggest that the proposed unified architecture metamodel can function as an effective interface, enhancing the accuracy, stability, and repeatability of LLM outputs.
However, the research also identified certain areas for improvement. The selected set of architectural diagrams requires optimization to eliminate redundancy and enhance clarity. Additionally, some diagrams need to be updated to incorporate further contextual orchestration, which is essential for a comprehensive understanding of the system architecture.
Conclusion
This work illustrates measurable advancements in the development of intelligent tools that automate SDLC processes. By presenting a comprehensive architecture that aligns with AI-driven development, the proposed unified architecture metamodel sets the stage for the next generation of software engineering practices. As organizations increasingly adopt AI technologies, the need for a cohesive architectural framework becomes paramount, ensuring that systems are robust, adaptable, and capable of meeting evolving business demands.