Document intake prep

When files arrive for annotation, a Beam agent reads each document's type and page count and registers it into the correct V7 Go project. It applies your intake rule to set the annotation template and priority, writing those fields back where the connector permits. Standard documents queue automatically for auto-annotation. When a file is unreadable, an unexpected format, or larger than your set limit, the agent holds it and notifies the data-prep owner. A person resolves the exception, so no malformed input silently enters a labeling project or skews the model's training data later.

Auto-annotation confidence routing

As V7 Go returns auto-annotations, a Beam agent reads the output and any confidence values attached to each page. It applies your threshold rule: high-confidence pages advance in the project while lower-confidence results are held for human labeling. The agent updates each item's status and writes the reason back to the task. Pages the model marks uncertain, or that conflict with an earlier label, go to a reviewer rather than straight through. Your annotators keep control over the hard cases, and the agent records which pages it passed and which it escalated.

Extraction hand-off to systems

Once a document is annotated, a Beam agent reads the extracted fields from V7 Go and matches them to a destination record in your connected system. It applies your rule for required fields, then writes the values across and updates a processing status where permissions allow. Complete, high-confidence extractions post without review. When a required field is empty, fails a format check, or the confidence is low, the agent routes that document to a person with the values it read. Someone corrects or confirms the data before it lands in a downstream system of record.