Сохранить в:
| Главные авторы: | , , |
|---|---|
| Формат: | Recurso digital |
| Язык: | английский |
| Опубликовано: |
Zenodo
2025
|
| Предметы: | |
| Online-ссылка: | https://doi.org/10.5281/zenodo.17839545 |
| Метки: |
Добавить метку
Нет меток, Требуется 1-ая метка записи!
|
Оглавление:
- <p>This paper introduces a prompt-based no-code<br>website generation platform that leverages retrieval-augmented<br>generation (RAG) with Google Gemini 2.5 Flash and visual<br>editing capabilities. The system integrates HNSWLib vector<br>storage with Xenova Transformers embeddings for context-aware<br>component retrieval, enabling efficient real-time code generation<br>from natural language prompts. Implemented with ReactJS (v19)<br>for the frontend, including Monaco Editor integration, and a<br>NodeJS/Express backend with JWT authentication, the platform<br>draws on a curated set of templates comprising standard UI<br>components and a full-stack e-commerce application. Evaluation<br>through automated testing demonstrates strong performance in<br>syntax correctness, functional completeness, generation success,<br>and RAG similarity, alongside effective e-commerce feature<br>detection across cart, pricing, and payment elements. These<br>results validate the use of RAG-enhanced prompt engineering<br>for accessible and scalable no-code web development tools.<br>Index Terms—Prompt-Based Website Generation, No-Code<br>Tools, Retrieval-Augmented Generation (RAG), Google Gemini<br>API, HNSWLib, ReactJS 19, NodeJS, Express, Visual Editing,<br>Web Development Automation</p>