Old vs. New Prompt Structure for Document Processor Flows

Users can create Document Processor Flows (DPFs) using either the old (legacy) or new prompt structure.

Old Prompt Structure

  • Follows a legacy design.
  • Supports Entity Classification and Document Classification independently.
  • Document Classification applies only to the first page, and the same class is assigned to the entire document.
  • The classification prompt is read-only and cannot be customized.
  • Entity and Table extraction prompt includes a list of entities and columns with clues separately.

New Prompt Structure

  • Enforces Document Classification before Entity Classification to ensure extraction is performed only on valid documents.
  • Page-level classification allows identifying and grouping pages by document type such as invoices and purchase orders within a single PDF file.
  • The classification prompt is editable, allowing users to add custom instructions for classification and boundary detection.
  • Refined extraction guidelines for Entity and table prompts.
  • Added a provision to add QA instructions.
  • Removed redundant prompts.
  • Updated User role definition for better accuracy.