This repository demonstrates how to "vibe code" a full-stack invoice processing application using Cursor/Claude Code from a simple starting file. This showcases building a complete Streamlit application with LlamaCloud's document extraction capabilities.

This repository provides a complete end-to-end solution for building intelligent document processing applications:

1. Vibe Coding Templates to Create Full Web Applications: Provides comprehensive prompts that can be used alongside coding assistants like cursor to one-shot transform simple scripts into production web applications
2. LlamaCloud Integration: Automatically creates and configures LlamaExtract extraction agents using predefined Pydantic schemas for structured data extraction
3. Sample scripts: Sample scripts that can be used as context for coding assistants while generating new streamlit apps!

This repository demonstrates vibe coding - rapidly building applications through iterative AI-assisted development. The goal is to transform a simple script into a full-featured application using Cursor/Claude Code.

• Python 3.11+
• LlamaCloud account and API key (Get one here)

git clone <your-repo-url>
cd invoice-extraction-vibe-coding
pip install -r requirements.txt

1. Go to LlamaCloud and log in
2. Create a new project and note your project_id and organization_id
3. Get your API key from your account settings

Copy the template and add your credentials:

cp .env.template .env

Then edit .env with your LlamaCloud credentials:

LLAMA_CLOUD_API_KEY=your_actual_api_key
LLAMA_CLOUD_PROJECT_ID=your_actual_project_id
LLAMA_CLOUD_ORGANIZATION_ID=your_actual_organization_id
LLAMA_CLOUD_AGENT_NAME=invoice_extraction_agent

python create_agent.py

This automatically creates an extraction agent in LlamaCloud using the invoice schema from sample_data/sample_schema.py.

python sample.py

This validates your configuration and runs extraction on the sample invoice. You should see structured JSON output.

Now comes the fun part - use the cursor_prompt.md template to transform sample.py into a full application:

1. Open your project in Cursor or Claude Code
2. Copy the prompt from cursor_prompt.md
3. Start iterating: "Transform this simple script into a professional Streamlit app with..."
4. Build features incrementally through natural language prompts

When you're done vibe coding, compare your creation with our app.py to see different approaches!

invoice-extraction-vibe-coding/
├── app.py                      # Full-featured Streamlit application (generated)
├── sample.py                   # Starting point - simple extraction script
├── create_agent.py            # Script to create LlamaCloud extraction agent
├── .env.template              # Environment variables template
├── requirements.txt           # Python dependencies
├── cursor_prompt.md           # Cursor prompt template for vibe coding
└── sample_data/               # Sample invoice data
    ├── batch1-0274.jpg       # Sample invoice image
    └── sample_schema.py      # Pydantic models for data structure

• Starting Point: sample.py - A basic 20-line script that calls LlamaCloud extraction
• End Result: app.py - A fully-featured Streamlit application with:
  • Professional UI with custom CSS
  • File upload functionality
  • Real-time data extraction
  • Structured data display
  • Invoice history tracking
  • Status indicators and metrics

• Document parsing and extraction using LlamaExtract
• Structured data output using predefined schemas
• Integration with kaggle_invoice_agent

This project demonstrates vibe coding - rapidly building applications through conversational AI development.

The cursor_prompt.md file contains a comprehensive prompt template that can potentially one-shot the entire application transformation:

1. Copy the main prompt from cursor_prompt.md
2. Paste it into Cursor/Claude Code with your sample.py file open
3. Watch the magic happen - the AI may build the entire Streamlit app in one go!

• Start Simple: sample.py - a working 20-line script
• Use Natural Language: Describe what you want, not how to build it
• Iterate if Needed: Add features through follow-up prompts
• Build Incrementally: Test and refine each addition

• Comprehensive Scope: Covers UI, functionality, and user experience
• Clear Context: References your actual files and sample data
• Specific Examples: Shows exactly what features to build
• Production-Ready: Asks for professional-quality output

Try the full prompt first - you might be surprised how much gets built in a single interaction!

• 📄 Multi-format Support: JPG, JPEG, PNG, BMP, TIFF
• 🎨 Professional UI: Custom CSS styling and responsive design
• 📊 Data Visualization: Structured invoice data display
• 📈 Analytics: Processing metrics and status tracking
• 🔄 Real-time Processing: Live extraction with progress indicators
• 💾 Session Storage: Invoice history within session

The application extracts structured invoice data including:

• Invoice metadata (number, date)
• Seller and client information
• Line items with pricing details
• VAT calculations and summaries
• Total amounts

See sample_data/sample_schema.py for the complete Pydantic model definitions.

1. Upload Invoice: Drag and drop or select an invoice image
2. Extract Data: Click the extraction button to process with LlamaCloud
3. View Results: See structured data in formatted tables and JSON
4. Track History: View processed invoices in the history tab

Update these values in your application for your own LlamaCloud setup:

project_id = "your-project-id"
organization_id = "your-organization-id"
agent_name = "your-agent-name"

• API Key: Ensure your LlamaCloud API key is valid and in .env
• Dependencies: Run pip install -r requirements.txt
• File Formats: Only image formats are supported
• File Size: Keep images under 10MB for best performance

This is an educational example demonstrating the vibe coding approach. Feel free to:

• Fork and modify for your use case
• Share improvements and variations
• Use as a starting point for your own projects

Open source - use freely for learning and development.

Built with ❤️ using LlamaIndex, LlamaCloud, and the power of vibe coding with Cursor/Claude Code!