Gemini JSON OCR

Gemini JSON OCR is a proof of concept showing how easy it is to use the latest Google Gemini to extract structured JSONs from documents.

Usage

$ export GOOGLE_API_KEY=<get your API key at https://aistudio.google.com/app/apikey>
$ uv run scan.py /Users/maurycy/Desktop/test
INFO:root:Processing file: MX-C304W_16122024_143019.pdf
Results for MX-C304W_16122024_143019.pdf have been written to /Users/maurycy/Desktop/test/MX-C304W_16122024_143019.pdf.json

Resulting in a JSON, such as:

{
  "waybill": {
    "scac": "SEAU",
    "booking_no": "4803804131",
    "bl_no": "4803804131",
    "vessel": "MERIDIAN",
    "containers": [ "TLLU5242619", "MSKU829454" ]
  }
}

Getting Started

Make sure that you've got uv installed:

# macOS
brew install uv
# macOS or Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
# Windows
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

(No need to install Python etc. uv will take care of that!)

Advanced usage

The prompt is in the prompt.txt.

Supported environment variables:

• GEMINI_MODEL, by default gemini-2.0-flash-exp
• GOOGLE_API_KEY, to be retrieved from Google AI Studio