Unified Architecture Metamodel of Information Systems Developed by Generative AI
New framework creates consistent 'Code to Documentation to Code' transformations across business, system, and developer layers.
Researchers Oleg Grynets and Vasyl Lyashkevych have published a paper proposing a Unified Architecture Metamodel specifically designed for information systems developed by generative AI. The core problem they address is the fragmentation that occurs when large language models (LLMs) automatically generate code and documentation without a consistent architectural framework. Their solution creates a structured interface between human architects and AI models, organizing system representation across three main layers: high-layer (business/domain understanding), middle-layer (system architecture), and low-layer (developer architecture).
The framework enables a closed transformation cycle they describe as 'Code to Documentation to Code,' where architectural diagrams provide consistent context for LLM generation. Their experiments demonstrated that using this structured architectural context significantly improves the accuracy, stability, and repeatability of AI-generated outputs. While effective, the researchers note the diagram set needs optimization to avoid redundancy and should be updated to better represent contextual orchestration. This work represents a measurable step toward intelligent tools that can automate more of the software development lifecycle (SDLC) while maintaining architectural coherence.
- Proposes three-layer architecture (business, system, developer) with specific diagrams for each layer to structure LLM context
- Enables 'Code to Documentation to Code' transformation cycle, improving generation stability and repeatability in experiments
- Serves as an interface between humans and AI models to reduce fragmentation in AI-developed systems
Why It Matters
Provides a blueprint for consistent, enterprise-grade software development using AI, moving beyond fragmented code generation.