AI Transaction Detection

STUB	This article is a stub. It contains minimal information on the topic and should be expanded.

The following is a placeholder taking from the Grooper Release Notes. More information on this topic can be found on your Grooper install's Help page.

AI Transaction Detection: New LLM-enabled Section Extract Method

• A “transaction” is a repeating data structure on a document, such as employee records in payroll reports or claims in an EOB.
• This is a specialized AI-powered section extract method designed for documents containing multiple, similarly structured transactions that may not be separated by explicit page breaks or static delimiters.
• AI Transaction Detection’s core is its “anchor-based boundary detection”.
  • The LLM is presented with a “detection window” (the starting N pages of the document) and is asked to identify a set of “anchor” features.
  • Anchors could be static text labels, regular expressions or other structured patterns that reliably indicate where each transaction starts.
  • The anchors are then matched against each line of the document, and each line is given a score based on how many anchors are matched. Lines must meet a minimum threshold to be considered a boundary point.
  • For each detected boundary, a new section instance is created, representing a single transaction.
• AI Transaction Detection differs from AI Collection Reader in how section boundaries are detected and then extracted.
  • AI Collection Reader splits the document into chunks of N pages and runs extraction on each chunk.
    • Issues can occur where sections span pages.
    • Works for a wide variety of documents.
  • AI Transaction Detection splits the document into transactions, then runs extraction on each transaction.
    • Handles sections that span pages well.
    • Specialized for transaction-based documents.