Technical Research Paper

Multi-Agent Systems for
Automated Design Generation

This research outlines the use of decoupled autonomous agents within Azure AI Foundry to solve visual consistency and technical adherence for MSMEs.

Authors: Diego Cancho | Jose Arenas | Lenis Wong
Institution: Universidad Peruana de Ciencias Aplicadas
Model: GPT-4o

Paper Title: GPT-4o Based Multiagent Web System For Automated Generation Of Cross Platform Design Systems for Micro, Small and Medium-sized Enterprises

Abstract

Micro, Small and Medium-sized Enterprises (MSMEs) face digital transformation limitations due to lack of resources and design teams. Only 1% fully automates their design system, and 52% report visual consistency issues. In Peru, although 93% of businesses promote digital transformation, 38% of workers lack necessary digital skills. Design Foundry is proposed as a GPT-4o multi-agent web system for automated cross-platform design system generation for MSMEs. The approach comprises three stages: (i) multimodal model benchmarking, (ii) GPT-4o agent implementation through Microsoft Foundry using three-agent architecture, and (iii) web application development. The system was validated using an MSME software product. Results show Design Foundry generated a complete design system including color palette, typography, reusable components, and exports to multiple programming languages and frameworks in approximately 40 minutes through natural conversation, compared to weeks required by manual design, maintaining professional visual consistency and facilitating direct development implementation.