Parse invoices and receipts

This is a short demonstration of how you can parse an invoice or receipt in Cradl. First, choose which method you would like to use for document parsing.

Using the Web UI

Locate the pre-trained models in the model overview and select the one matching your use case. You may upload your own document or use one of our sample documents for testing.

• For invoice parsing click here
• For receipt parsing click here

Using the CLI

info

You must have the CLI installed. Follow the instructions here to install the CLI.

To parse an invoice in Cradl, you need to make two API calls. One to upload a document and a second call to parse the document, creating a prediction. The second API call requires you to provide a modelId. If you are using one of our pre-trained models they are las:organization:cradl/las:model:invoice and las:organization:cradl/las:model:receipt for the invoice and receipt models, respectively. You can use these as a quick test, but we recommend training your own customized model for production usage. Read more about how to train a high accuracy model here

las documents create <your invoice.pdf>
las predictions create <documentId> las:organization:cradl/las:model:invoice

Congratulations, you have now successfully parsed your first invoice using Cradl. The JSON response body should have a list of predictions for your uploaded document. The JSON payload should look similar to the following.

{
  "predictionId": "las:prediction:...",
  "documentId": "las:document:...",
  "modelId": "las:organization:cradl/las:model:invoice",
  "predictions": [
    {
      "label": "vat_amount",
      "page": 1,
      "value": "75.72",
      "confidence": 0.9975606850585552
    },
    {
      "label": "total_amount",
      "page": 0,
      "value": "369.00",
      "confidence": 0.9931743085807881
    }
  ]
}

Using Postman

tip

Check out our Postman Collection here.

Log in to Cradl and create API credentials

First of all, log in to Cradl and create an app client and download the file containing client ID and client secret. These values are needed for authorization when using the API.

Configure authorization in Postman

Go to our postman page and locate our API listed under the "APIs" tab. There should only be one API there named "Cradl". Fork the Collection and navigate to the Authorization settings of the collection as shown below.

Edit the Client ID and Client Secret and add the values for client ID and client secret that you downloaded in the Cradl app from the previous step. When you are done adding your client ID and secret, click on the "Get New Access Token" button and validate that you are getting an "Access Token". Then click the "Use Token" button in the dialog that appears.

Upload and parse an image or PDF invoice

In order to parse an invoice in Cradl we need to make two API calls, one that uploads a document and another that parses it and creates predictions.

To upload a document we need to invoke POST /documents with a JSON payload containing the image or PDF as a base64 encoded string together with its content mime type. Cradl supports image/jpeg, image/png, image/tiff and application/pdf content types. Since there is no easy way of converting a file to base64 directly in Postman, we'll show how you can achieve this using the base64 program on linux.

base64 <invoice filename> -w 0

Copy the output and paste it in the JSON body as the value of the key "content" and put the corresponding mime type as value for the key "contentType", ignore all other parameters for now. Note the documentId from the API JSON response body as we'll need this value for the next part.

info

If you get an "Unauthorized" response from the API, click on the "Authorization" tab of the request. "Type" should be "Inherit auth from parent", but due to a limitation in Postman we'll have to reselect "Inherit auth from parent" for the request to update headers from configuring Client ID and secret earlier.

Finally, we are ready to create predictions on our invoice and to do that you'll need to invoke POST /predictions with documentId from the previous step and with a modelId. The modelId determines which model we are using to create predictions. Cradl offers several out-of-the-box models that you can use, and they have deterministic values for modelId, namely las:organization:cradl/las:model:invoice and las:organization:cradl/las:model:receipt. As with the previous API request, ignore all other parameters in the JSON request body.

Congratulations, you have now successfully parsed your first invoice using Cradl. The JSON response body should have a list of predictions on your uploaded document and the payload should look similar to the following.

{
  "predictionId": "las:prediction:...",
  "documentId": "las:document:...",
  "modelId": "las:organization:cradl/las:model:invoice",
  "predictions": [
    {
      "label": "vat_amount",
      "page": 1,
      "value": "75.72",
      "confidence": 0.9975606850585552
    },
    {
      "label": "total_amount",
      "page": 0,
      "value": "369.00",
      "confidence": 0.9931743085807881
    }
  ]
}