Node.JS and Azure Cognitive Services

Handy solutions Java script - Node.JS
,
1

Simple example on how you can use Node.JS processor with connection to Azure cognitive services especially Form Recogniser. Before going further review the Form Recogniser service description and use: Invoice data extraction – Document Intelligence (formerly Form Recognizer) - Azure AI services | Microsoft Learn

In my case I want to use prebuilt model as well custom model for additional fields on the invoice. I want to use same endpoint for getting the data from invoices (invoice-prebuilt + custom) and generic document with model prebuilt-document.

Attachment is stored in Xeelo so in the first place I download the attachment content using Xeelo Attachment Download connector. Then I pass the data into Azure services. I want to run the data extraction for prebuilt-invoice and custom model in parallel so I use const responseAll = await Promoise.all([promise1, promise2]); where promises are coming from my own function that posts the data in Azure and in loop with one-second delay trying to get the results.

Azure credentials (key) as well as URL of endpoint under my subscription are stored in general variables as sensitive data so I do not expose them by accident.

// functions

async function getOCRData(stringURL){

  const Key =  "${#AzureOCR_Key}"

  // upload file to Azure OCR
  const responsePOST = await fetch(stringURL, {
    method: "POST",
    body: JSON.stringify({"base64Source": inputData[0].DocumentContent}),
    //body: JSON.stringify({"urlSourc1e": DocumentUrl}), 
    headers: {
      "Content-Type": "application/json", 
      "Ocp-Apim-Subscription-Key": Key
      }
  });

  if(!responsePOST.ok){
    log.error("Upload of document into Azure OCR was not successful with following error:");
    const resp = await responsePOST.json();
    log.error(JSON.stringify(resp.error.message));
    log.error(JSON.stringify(resp.error?.innererror.message));
  }

  // extract call back url from document upload
  const CallbackURL = responsePOST.headers.get("Operation-Location");

  var OCRDataStatus = "new";
  var OCRData = null;

  // get document OCR in cycle with 3 second delay
  while(OCRDataStatus != "succeeded"){

    const responseGET = await fetch(CallbackURL, {
      method: "GET",
      headers: {
        "Ocp-Apim-Subscription-Key": Key
        }
    });

    if(!responseGET.ok){
      log.error("Document fetch from Azure OCR was not successful.")
    }

    OCRData = await responseGET.json();
    OCRDataStatus = OCRData.status;

    if(OCRDataStatus != 'succeeded'){
      log.warn("Azure OCR status: " + OCRDataStatus);
      await sleep(1000);
    };

  };

  return OCRData;

}

// main code

const EndpointURL = "${#AzureOCR_URL}";
const ModelID = inputData[0].ModelID;

var OCRData = {}

if(ModelID == "prebuilt-invoice"){

  const AzureURL = EndpointURL + "/formrecognizer/documentModels/prebuilt-invoice:analyze?api-version=2023-07-31&features=keyValuePairs"
  const OCRDataPromise = getOCRData(AzureURL);

  const AzureURLExt = EndpointURL + "/formrecognizer/documentModels/Invoice_Extended_v12:analyze?api-version=2023-07-31"
  const OCRData2Promise = getOCRData(AzureURLExt);

  const OCRDataResult = await Promise.all([OCRDataPromise, OCRData2Promise]);

  OCRData = OCRDataResult[0];
  const OCRData2 = OCRDataResult[1];

  if(OCRData2.analyzeResult.documents?.length > 0){
    OCRData.analyzeResult.documents.push(...OCRData2.analyzeResult.documents)
  }

} else {

  const AzureURL = EndpointURL + "/formrecognizer/documentModels/prebuilt-document:analyze?api-version=2023-07-31"
  OCRData = await getOCRData(AzureURL);

}

return [{
  "JSON": OCRData
}];
5 Likes