Form Recognizer Solution Accelerator

Accelerate your Form Recognizer solution to production with this Solution Accelerator, which leverages an Azure Function and a set of Logic Apps to split multi-page PDF files to single-page PDF files and sends individual PDF files to the REST API endpoint of a trained custom document model in Form Recognizer.

This solution implements two capabilities that are commonly required when working with a trained custom document model:

1. Splitting multi-page PDF documents into individual, single-page PDF documents
2. Analyzing the results of documents sent to the Form Recognizer REST API endpoint of a trained custom document model

Please reference this blog post for detailed, step-by-step instructions for how to implement this solution. We are also actively working on organizing the same step-by-step instructions in this repository.

Step 1: Deploy core resources to Azure

Using the below button, six Azure services will be deployed:

• Storage account
• Function app
• App Service plan
• Form Recognizer
• Logic app (x2)

Step 2: Create containers & upload data

Download sample data from this repository and upload it into the new containers you create.

Step 3: Train custom document model

Open the Form Recognizer Studio and train a custom document model.

Step 4: Deploy open-source Python code to split PDFs

Deploy open-source Python code to your Function App to split multi-page PDF files.

Step 5: Configure Logic App to split multi-page PDF documents to single-page PDF documents

Create a Logic App to call your Azure Function App and save individual PDF files based on a multi-page PDF file input.

Step 6: Configure Logic App to send single-page PDF document data to REST API endpoint of trained custom document model

Leverage the REST API endpoint of a trained custom document model in Form Recognizer.

Step 7: Verify the results

Upload a multi-page PDF file and verify that the first Logic App produces single-page PDF files. Then, verify that the second Logic App sends each file to the custom model endpoint in Form Recognizer and saves the resulting JSON.