🛠️ LangGraph App Builder

An AI-powered application generator that converts natural language descriptions into functional projects using a multi-agent LangGraph architecture.

---

🏗️ Architecture

• Planning Agent – Creates a project plan with tech stack and features
• Architect Agent – Breaks down the plan into detailed implementation tasks
• Coding Agent – Executes tasks and generates actual code files

---

🚀 Quick Start

Prerequisites

• Python 3.11+
• Google Gemini API key
• uv package manager

Installation

# Create virtual environment
uv venv
.venv\Scripts\activate  # Windows
source .venv/bin/activate  # Linux/Mac

# Install dependencies
uv pip install -e .

# Setup environment
cp .sample_env .env
# Add your GEMINI_API_KEY to .env

Running the App

python agent/graph.py

Enter your project description when prompted. Examples:

• "Create a simple login page with colorful design"
• "Build a to-do list app using HTML, CSS, and JavaScript"
• "Create a FastAPI blog with SQLite database"

---

📁 Project Structure

agent/
  ├── graph.py      # Main LangGraph orchestration
  ├── prompts.py    # Agent prompts
  ├── states.py     # Pydantic models
  ├── tools.py      # File I/O tools
  └── visualize.py  # Graph visualization
main.py            # Entry point
pyproject.toml     # Dependencies

⚙️ Configuration

Key settings in agent/graph.py:

llm = ChatGoogleGenerativeAI(
    model="gemini-2.0-flash-exp",
    temperature=0.2
)

tested opensource models with limited use because of Tokens per minute limit:

• openai/gpt-oss-20b
• gemini-2.0-flash
• gemini-1.5-pro (higher accuracy, slower)

---

📸 Example Projects

1. Calculator Web App

Generated simple calculator with HTML, CSS, and JavaScript.

2. Login Page

Beautiful login interface with responsive design.

3. Portfolio Website

Personal portfolio with multiple sections and responsive layout.

---

🐛 Troubleshooting

Issue: "Tool choice is required" error

• Solution: Model does not support structured outputs. Try a different model with better tool-use support.

Issue: API quota exceeded

• Solution: Google Gemini has rate limits. Wait before retrying or use a different model.

Issue: Generated project won't run

• Solution: Check generated_project/README.md for specific setup instructions.

---

---