How the solution works

You can customize the way this solution processes documents. The solution orchestrates ML inferences, using AWS AI services and their pre-trained models, to extract content and automate processing. The solution’s XML-based configuration provides customization for the orchestration workflow for respective use cases.

Workflows

As shown in the architecture diagram, the solution deploys three workflows: text extraction, entity detection, and redaction. The workflow orchestrator Lambda function orchestrates the order and method of processing uploaded documents using any of these workflows.

Workflow configurations

The solution stores its application workflow configurations in the workflow-config DynamoDB table. The solution uses these workflow configurations to:

• Set the number and details of documents required to be uploaded to start processing

• Set the workflows that each document needs to be processed with

The solution creates these DynamoDB table records when the application is deployed. To create the tables, the solution uses configuration JSON files available in the workflow-config directory at the root of the application source code.

To add a new configuration, you can clone an existing record in the workflow-config DynamoDB table. You can add these records after deployment by signing in to the DynamoDB Console. The format of this configuration file is described below.

You can select which workflow configuration to use by setting the WorkflowConfigName parameter during deployment. The workflow orchestrator Lambda function uses this parameter input as the key to retrieve the desired configuration form the workflow-config DynamoDB table. This parameter has a default value of default.

The following JSON object shows a sample workflow configuration file. During deployment, the solution serializes configuration files such as these into DynamoDB record data and added to the workflow-config table. The key of the table corresponds to the Name parameter of the configuration file.

{
 "Name": "textractToEntity",
 "WorkflowSequence": [
   "textract",
   "entity-standard"
  ],
 "MinRequiredDocuments": [
    {
     "DocumentType": "generic",
     "FileTypes": [
       ".pdf",
       ".png",
       ".jpeg",
       ".jpg"
      ],
     "RunAmazon TextractAnalyzeAction": false,
     "MaxSize": "5",
     "WorkflowsToProcess": [
       "textract"
      ]
    },
    {
     "DocumentType": "receipt",
     "FileTypes": [
       ".pdf",
       ".png",
       ".jpeg",
       ".jpg"
      ],
     "RunAmazon TextractAnalyzeAction": true,
     "AnalyzeDocFeatureType": ["TABLES", "FORMS", "SIGNATURES"],
     "MaxSize": "5",
     "WorkflowsToProcess": [
       "entity-standard"
      ]
    }
  ]
}

The following table describes the details of the configuration.

Parameter	Type	Description	Supported values
Name	String	Name of the workflow configuration. This corresponds to the WorkflowConfigName CloudFormation parameter required during deployment.	Any
WorkflowSequence	Array<String>	The sequence of the document processing workflows to run on an uploaded document, in the order described. Note The solution follows the order of items in this list.	textract entity-standard entity-pii entity-medical redaction
MinRequiredDocuments	Array<Map>	This list map object describes the types of documents along with their specs required to execute this workflow. The number of items in this list indicate the number of documents required. The details of the map are described in the following section.	See the following table

The MinRequiredDocuments parameter in the configuration file is a list of the required documents to create a workflow. Each item in this list corresponds to the configuration of a single document. Workflow processing starts only once all of the required types of documents are uploaded to a case.

Parameter	Type	Description	Supported values
DocumentType	String	The user-ascertained type of the uploaded document. Based on the document type, the solution runs the corresponding Amazon Textract analyze action. There are three textract actions to analyze a document: AnalyzeID for identity documents AnalyzeExpense for expense-related documents, such as receipts AnalyzeDocument for generic documents from which key-value pairs are extracted	This solution supports the following types of documents driving-license passport receipt invoice vaccination-card paystub loan-information health-insurance-card generic (any other type not listed above)
FileTypes	Array<string>	File type of the uploaded document	jpeg jpg png tiff pdf
MaxSize	Integer	Maximum file size in megabytes (MB) of a single uploaded document. Note The solution has a maximum page limit of 15 pages. This limit is set to ensure that synchronous Amazon Textract operations can run reliably. Based on your use case, you may choose to customize the limit. Doing so may impact the system’s reliability to handle larger file sizes or additional pages.	Up to 5 MBM
WorkflowsToProcess	Array<string>	This list is a subset of the WorkflowSequence parameter described in the previous table. It indicates the type of processing to run on a specific type of document. You can use this parameter to have fine-grained control of the orchestration process.	textract entity-standard entity-pii entity-medical redaction
RunTextractAnalyzeAction	(Optional) Boolean	If you set this parameter to true, it Amazon Textract AnalyzeDocument and DetectDocumentText runs for the document.	true false
AnalyzeDocFeatureType	(Optional) Array<string>	The type of features to detect when Amazon Textract AnalyzeDocument is called. If you set RunTextractAnalyzeAction to true but this value is missing, then the solution uses a default value of ["TABLES", "FORMS", "SIGNATURES"]. For more information, see AnalyzeDocument in the Amazon Textract Developer Guide.	List containing any of: TABLES FORMS SIGNATURES