A Novel Framework for Intuitive E-commerce Design: Leveraging Large Language Models to Translate User Commands into Functional Code

Espinoza N.; Soto A.; Navarrete E. · Proceedings of the International Conference on Soft Computing and Machine Intelligence, ISCMI (2025) · DOI: 10.1109/iscmi67495.2025.11358555

Resumen

This study evaluates the performance of leading Large Language Models for generating front-end code for e-commerce applications from natural language commands, aiming to lower technical barriers for entrepreneurs. The models analyzed include GPT-4.5, Llama 4 Maverick, Claude 4 Sonnet, and Mistral Medium 3. These models were benchmarked on a standardized set of 10 front-end generation tasks under controlled conditions, with fixed system prompts and generation parameters. The evaluation employed a multi-faceted suite of metrics including Functional Success Rate, Structural Adherence Score, Code Quality Score, and a Time-Cost analysis. While proprietary models achieved the highest quality scores, Llama 4 Maverick emerged as the most suitable model, providing the best balance between robust performance and operational efficiency. Based on these findings, this study proposes the integration of the mentioned model into a no-code web application, offering a practical and accessible tool for startups to build digital stores.

Instituciones: Universidad Peruana de Ciencias Aplicadas