Reproduce “MAAD: Automate Software Architecture Design through Knowledge-Driven Multi-Agent Collaboration” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ruiyin Li, Yiran Zhang, Xiyu Zhou, Peng Liang, Weisong Sun, Jifeng Xuan, Zhi Jin, Yang Liu
The MAAD framework introduces a knowledge-driven multi-agent system to automate software architecture design, generating comprehensive architectural blueprints and evaluation reports directly from Software Requirements Specifications. It demonstrates superior performance over general-purpose multi-agent baselines by providing finer artifact granularity and integrating systematic architecture evaluation.
Key insights
- Specialized multi-agent systems with clearly defined roles are more effective for complex, knowledge-intensive tasks like software architecture design than general-purpose LLM-based systems.
- Integrating external knowledge qualitatively improves aspects of architectural design (e.g., modularity), but its impact on quantitative mismatch rates is limited without more domain-specific knowledge.
- The choice of the foundational LLM significantly affects the quality and consistency of generated architecture designs, indicating a need for task-aware LLM evaluation and selection for specific agent roles.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.