They use the OCR + structured data extraction in this example, but I'm thinking you could extrapolate this to any scenario where you are extracting some data (from a table, markdown file, OCRed doc), and then use a decision model to perform a fast on-the-fly review/validation.