A Methodology for Generating Mobile Applications Through Large Language Models

This repository contains the supplementary materials for our paper, titled "A Methodology for Generating Mobile Applications Through Large Language Models". In this work, we present and validate a platform-independent method for generating and evaluating complete mobile applications from high-level natural language descriptions. Our approach utilizes a multi-agent Large Language Model (LLM) framework to manage a three-stage process: App Description Generation, Code Generation, and Evaluation. 📂 Repository Structure This repository is organized into three main directories, corresponding to the stages of our methodology: ├── 1.app-description/ # Natural language app descriptions │ ├── boobie-goods.txt │ ├── duolingo.txt │ └── threads.txt ├── 2.app-generation/ # Generated mobile applications │ ├── boobie-goods.zip │ ├── duolingo.zip │ └── threads.zip ├── 3.evaluation/ # Research evaluation materials │ ├── boobie-goods.pdf │ ├── duolingo.pdf │ ├── threads.pdf │ └── legend.png └── README.md # This file 📝 How to Use This Repository 1. App Description The 1.app-description/ directory contains the detailed mobile app specification documents generated by the "product manager" LLM agent. These .txt files served as the blueprint for the code generation phase. Each file includes: App purpose: The main objective of the application. Detailed Feature List: A comprehensive list of functionalities. Layout and components: A description of the app's screens and UI elements. User interaction flow: A complete map of possible user journeys within the app. 2. App Generation The 2.app-generation/ directory contains the complete source code for the three mobile applications generated by our methodology. The applications were developed using the Flutter framework. To use these files: Unzip the desired application file (e.g., unzip boobie-goods.zip). Navigate into the project directory. Follow the standard procedure to build and run a Flutter application. 3. Evaluation The 3.evaluation/ directory contains the reports from our code quality assessment, which was performed using SonarQube. Each PDF file provides a detailed static analysis of the generated source code, focusing on: Reliability (Bugs) Security (Vulnerabilities) Maintainability (Code Smells) Duplications Complexity The legend.png file provides a key to understanding the ratings and metrics used in the SonarQube reports.本仓库包含我们题为《基于大语言模型生成移动应用的方法论》的论文补充材料。本研究提出并验证了一种平台无关的方法，可从高层次自然语言描述生成并评估完整的移动应用程序。我们的方案采用多智能体大语言模型（Large Language Model，LLM）框架，管理三阶段流程：应用描述生成、代码生成与评估。 📂 仓库组织结构 本仓库分为三个核心目录，分别对应本研究方法的三个流程阶段： ├── 1. 应用描述/ # 自然语言应用描述文件 │ ├── boobie-goods.txt │ ├── duolingo.txt │ └── threads.txt ├── 2. 应用生成/ # 生成的移动应用程序 │ ├── boobie-goods.zip │ ├── duolingo.zip │ └── threads.zip ├── 3. 评估/ # 研究评估材料 │ ├── boobie-goods.pdf │ ├── duolingo.pdf │ ├── threads.pdf │ └── legend.png └── README.md # 本说明文件 📝 使用指南 1. 应用描述 1.app-description/ 目录存放由“产品经理”AI智能体（AI Agent）生成的详细移动应用规格文档，这些.txt文件作为代码生成阶段的蓝图。每份文件包含以下内容： - 应用用途：应用的核心目标 - 详细功能列表：功能的完整清单 - 布局与组件：应用界面与UI元素的说明 - 用户交互流程：应用内所有可行用户路径的完整映射 2. 应用生成 2.app-generation/ 目录存放本方法生成的三款移动应用的完整源代码，所有应用均采用Flutter框架开发。 使用方法如下： 1. 解压目标应用文件（例如执行unzip boobie-goods.zip）； 2. 进入项目目录； 3. 按照标准流程构建并运行Flutter应用。 3. 评估 3.evaluation/ 目录存放我们使用SonarQube开展代码质量评估的报告。每份PDF文件均对生成的源代码进行详细静态分析，聚焦以下维度： - 可靠性（缺陷） - 安全性（漏洞） - 可维护性（代码异味） - 重复代码 - 复杂度 legend.png 文件用于说明SonarQube报告中的评级与指标含义。