Define the schema
Choose the fields, formats, and missing-value rules your downstream systems need.
USE CASE / DOCUMENT EXTRACTION
Messy documents. Consistent structure.
Define the fields that matter, collect representative documents, and measure whether outputs match your schema.
Explore the workspaceTHE STARTING POINT
A real workflow.
A focused experiment.
An operations team wants to reduce repeated manual entry from supplier invoices. A useful first experiment focuses on a small, stable schema before adding more document types.
ILLUSTRATIVE EXAMPLE · NOT A LIVE MODEL OUTPUT
Extract supplier, currency, total, and invoice date.
{
"supplier": "Northstar Supplies",
"currency": "EUR",
"total": 1240,
"invoice_date": "2026-10-01"
}THE WORKFLOW
Make the task repeatable.
Collect the difficult cases
Include unusual layouts, incomplete fields, and examples where the correct response is an empty value.
Compare field by field
Use expected and actual outputs to find formatting errors and incorrect values.
WHAT TO CHECK
Measure the details
that matter.
- Field completeness
- Exact matches for normalized values
- Valid JSON structure
- Handling of missing information
KEEP EXPLORING