demo3.md

Demo 3 - Creating a Custom Invoice Reader Skill

Summary

In this exercise we create an Azure Search Skill (in the form of an Azure Function) that takes data from Azure Search and returns a structured json document representing the contents of the pdf Invoice. To get a more in depth treatment of creating an Azure Function please refer to the docs. It is assumed that a storage account has been created (as described in the first demo as well as the Form Recognizer service created in the second demo)

What you need

• Invoice Test Set (used to test the Azure Function)

• Postman is used to send requests to the Form Recognizer service REST API. Refer to this short primer to learn more.

• Postman Reader Skill Request collection.

• Azure Functions Core Tools

Azure Resources

The table below describes the resources created throughout this exercise.

Name	Type	Purpose
readerskillstorage	Storage Account	Storage used for Azure Function
readerskill	Function App	Cognitive Skill App
readerskill	App Insights	Adds insights to Function App
EastUS2LinuxDynamicPlan	App Service Plan	Consumption based plan for running Function App (this is created implicitly)

What to do

There are three main steps:

1. Run the InvoiceReaderSkill Azure Function locally
2. Create resources to support the InvoiceReaderSkill Azure Function
3. Run the InvoiceReaderSkill on Azure

Run the Azure Function Locally

1. Download the InvoiceReaderSkill code found in this repo in Visual Studio Code (make sure the Azure Tools are installed).

2. Add a local.settings.json file to the folder with the following contents:

{
  "IsEncrypted": false,
  "Values": {
    "FUNCTIONS_WORKER_RUNTIME": "python",
    "AzureWebJobsStorage": "{AzureWebJobsStorage}",
    "FormsRecognizerEndpoint": "<YOUR_FORMS_ENDPOINT>.cognitiveservices.azure.com",
    "FormsRecognizerKey": "<YOUR_FORMS_KEY>",
    "ModelId": "<YOUR_TRAINED_MODEL_ID>"
  }
}

3. Run the func host start command to start the function.

4. Import the Postman Reader Skill Request collection to Postman

5. Open the Local Forms Skills Request and change the Pre-request script to set the storageAccount variable to your storage account (in the video the variable is ttinvoicestorage) and the SASValue to the appropriate Secure Access Signature to the train container. To learn how to get a Secure Access Signature, refer to our brief explanation.

pm.environment.set('storageAccount', '<YOUR STORAGE ACCOUNT>')
pm.environment.set('container', 'test')
pm.environment.set('SASValue', '<SAS>')

6. Verify that the POST uri matches the one for the Azure Function (in our running of it we use http://localhost:7071/api/AnalyzeInvoice) and click on Send. You should get a response similar to that below:

Create Azure Function Resources

It is assumed that the following commands are run in the same directory as the actual Function and you have logged into the Azure CLI using az login

1. Run the following Azure CLI Commands to create the appropriate resources and publish the function:

Create Function Storage

az storage account create --name {FUNCTION_STORAGE} --location {LOCATION} --resource-group {YOUR_RG} --sku Standard_LRS

Create Function App

az functionapp create --resource-group {YOUR_RG} --consumption-plan-location {LOCATION} --name {FUNCTION_NAME} --storage-account {FUNCTION_STORAGE} --runtime python --os-type linux

Publish Function

func azure functionapp publish {FUNCTION_NAME} --build remote

2. Use Visual Studio Code to sync the local settings to the Azure Function by selecting the Azure button, navigating to the newly created function, right clicking on Application Settings and selecting Upload Local Settings (NOTE: Do not overwrite the existing variables).

3. Browse to the newly created Function App in the portal to add Application Insights to the service.

Run the Azure Function in the Cloud

1. Open the Cloud Forms Skills Request and change the Pre-request script to set the storageAccount variable to your storage account (in the video the variable is ttinvoicestorage) and the SASValue to the appropriate Secure Access Signature to the train container. To learn how to get a Secure Access Signature, refer to our brief explanation.

pm.environment.set('storageAccount', '<YOUR STORAGE ACCOUNT>')
pm.environment.set('container', 'test')
pm.environment.set('SASValue', '<SAS>')

2. Verify that the POST uri matches the one for the Azure Function (should look something like https://<YOUR_ENDPOINT>.azurewebsites.net/api/AnalyzeInvoice) and click on Send. You should get a response similar to that below:

Next Demo

Learn how to tie everything together and finally build the end to end solution by continuing on to Demo 4 - Tying it all together