Applied AI · illustration
A document becoming a row.
This is a scripted illustration, not a live run. Nothing is uploaded, nothing is processed, and the values below are written into the page. It exists to show the shape of the work — a document on the left, the row a clerk would otherwise have typed on the right — and it makes no claim about how accurate the real thing is on your paperwork.
Commercial invoice
| HS Code | Description | Qty | Unit | Amount |
|---|---|---|---|---|
| 5208.52 | Cotton woven fabric, dyed | 4,000 m | USD 1.85 / m | USD 7,400.00 |
Terms: CIF Jebel Ali · Country of origin: India
Most pages are handled by the local steps and only the pages that genuinely need reading reach the model call. How that split falls on a real document set has not been measured, and the local steps still cost compute to run.
The read-out plays here. If it does not, the page still describes the workflow above and below — the animation is an illustration, not the argument.
Representative document · illustrative of the pipeline · not a live client run and not a measured result
What would make this a real claim
Your documents are not this one.
The only number worth anything here is how often the pipeline is wrong on your paperwork, and that number does not exist until someone hand-checks a set of your documents and scores against it. Deterministic checks catch particular classes of error — a total that does not add up, a date outside a plausible range — but catching those is not the same as being right, and I will not describe it as though it were.
Until that measurement exists for a given document type, treat this page as what it says it is: an illustration of the shape of the work.