Design Panda: AI-driven Optimal Software Architecture Artifact Generation

Authors:

S. Kanmani¹, Aaryan M², Syed Khizr Tahseen ³, T Daranish⁴

Professor, Information Technology, Puducherry Technological University, Puducherry, India ¹

Tech Student, Information Technology, Puducherry Technological University, Puducherry, India ^{2 3 4}

Abstract: Software architecture design is a critical phase in the software development lifecycle, influencing system scalability, maintainability, and performance. However, traditional architecture design relies heavily on manual effort and expert knowledge, making it time-consuming and prone to inconsistencies. This paper presents Design Panda, an AI-driven framework for automated software architecture artifact generation. The system leverages Natural Language Processing (NLP) to extract structured requirements from unstructured inputs and integrates Retrieval-Augmented Generation (RAG) to ground architectural decisions in established software engineering knowledge. A multi-agent system consisting of Architect, Critic, Optimizer, and Arbiter agents enables iterative generation, evaluation, and refinement of architecture designs. The proposed framework enhances transparency, reduces design errors, and produces validated architecture outputs including UML diagrams and documentation, making it suitable for modern intelligent software design systems.

Keywords: Software Architecture, Requirement Engineering, Natural Language Processing, Retrieval-Augmented Generation, Multi-Agent Systems.