WARD GEYS↖ Back to portfolio

CASE STUDY / FREIGHT DOCUMENT PARSERFROM PAPER TO REVIEWABLE DATA

Freight document parser

Read the paper.
Keep the check.

A freight-document prototype that extracts structured fields from a photo or upload, then gives a desk worker a chance to review and correct the result.

Platform
Browser · camera and upload workflow
My role
Capture/review workflow, validation decisions and AI-assisted implementation
Stage
In development · no live terminal integration
FREIGHT DOCUMENT PARSER / WORKFLOW VISUALIZATIONINSIDE THE IDEA
CMR / DOCUMENT

Freight information

Vehicle · Carrier · Destination

DESK / HUMAN REVIEW

Extracted.
Then checked.

Vehicle detailsReview
CarrierReview
DestinationReview

Edit the fields. Confirm the result.

A visualization of the project workflow, using illustrative content. This is not an application screenshot.

THE STARTING POINT

The next piece
of terminal friction.

A freight document holds information someone at a desk needs to enter elsewhere. The parser explores shortening that retyping step while keeping a person in the review loop.

It is related to my terminal-assistance work, but it is a separate prototype. Camera capture or file upload feeds a vision model, and the result returns as fields the desk can inspect.

HOW IT FITS TOGETHER

Inside
the workflow.

Keep capture, extraction and human review as distinct steps.

  1. 01

    Capture

    Photograph the document or upload a file. The capture view checks placement and image quality.

    CAMERA · FILE UPLOAD
  2. 02

    Extract

    Use a vision model to identify the document fields and vehicle information.

    VISION API · VALIDATION
  3. 03

    Review

    Flag uncertain results, allow corrections and prepare a structured result.

    HUMAN REVIEW · JSON

INSIDE THE BUILD

The parts
that matter.

Treat extraction as a proposal.

The model output is checked against a Zod schema and VIN validation. These checks catch structural and identifier problems; they do not prove that every field matches the paper document. The desk user can correct the fields before confirming the record.

Separate failure from confirmation.

Timeouts, retries and a manual fallback are part of the workflow. Confirmation is a separate step that produces the structured payload and an audit record. A model response alone is not a confirmed operational record.

Integration remains ahead.

The prototype prepares structured output. A working connection to a live terminal operating system is not demonstrated here.

Extract first.
Verify before use.

WHAT I TOOK FROM IT

The useful output is something a desk worker can check and correct. The model is one step in that workflow.

NEXT PROJECT / ATLAS STORY STUDIO

Another problem.
Another build.

Explore Atlas Story Studio