Invoice extractor

Scripts extracting data from the Czech invoices in JSON and checking validity of VAT tax payer. First script uses Anthropic Claude 3.5 Haiku and you need API key for its running. The second script uses Llama 4 Maverick provided by Groq and you need Groq API Key. For cheaper processing, script turns multipage invoices in PDF to individual images and only use the first invoice page. According to our intense testing of many types of invoices we achieved accuracy of extraction 92-93% for Haiku 3.5 model. If you would like to get better results with accuracy around 95% you would need to employ the more powerful model Claude Sonnet 4. To do so, you have to change one line of code with the end point of the Sonnet model. Using Llama 4 Maverick is with similar accuracy level like Sonnet 4.

Installation Instructions

Using Traditional Venv + Pip

On macOS/Linux:

# Create virtual environment
python3 -m venv invoice_env

# Activate virtual environment
source invoice_env/bin/activate

Install Dependencies

# Upgrade pip (optional but recommended)
pip install --upgrade pip

# Install required packages
pip install anthropic PyMuPDF Pillow

Place Your Files and Run

# Copy the invoice_extractor.py script to the project directory
# Copy your invoice files (PDF/images) to the same directory
# Edit the script to add your API key

# Run the script
python3 invoice_extractor.py

Deactivate Environment (When Done)

deactivate

Quick Setup Script

For Venv Users (macOS/Linux):

mkdir invoice-extractor && cd invoice-extractor
python3 -m venv invoice_env
source invoice_env/bin/activate
pip install anthropic PyMuPDF Pillow
# Copy your script and invoice files here
# Edit API key in the script
python invoice_extractor.py

---

Package Descriptions

• anthropic: Official Python client for the Anthropic API
• groq: Official Python client for the Groq API
• PyMuPDF: Library for working with PDF files (also known as fitz)
• Pillow: Python Imaging Library for handling various image formats

---

Llama 4 using Groq as inference provider

Create virtual environment for "my_project"

python3 -m venv my_project_venv

Activate venv

source my_project_venv/bin/activate  # Linux/Mac
# or
my_project_venv\Scripts\activate     # Windows

Install environment variables with Groq API Key

echo "GROQ_API_KEY=your_groq_key_here" > .env

Instal libraries and dependencies

pip install groq PyMuPDF Pillow requests

Troubleshooting

Common Issues:

1. Python not found: Make sure Python 3.8+ is installed
2. Permission errors on Windows: Run PowerShell as Administrator for UV installation
3. Virtual environment activation issues: Check your shell (bash, zsh, PowerShell)
4. Package installation fails: Try upgrading pip first: pip install --upgrade pip

Verifying Installation:

# Check Python version
python --version  # Should be 3.8 or higher

# Check if packages are installed (after activation)
python -c "import anthropic, fitz, PIL; print('All packages installed successfully!')"