Contact

Program case study

AgenticOCR

AI-assisted document-to-data workflows with human review.

How a document-to-data workflow keeps the source page beside the extracted result, and why the program is presented as a staging showcase rather than a production system.

Staging showcase — live

  • Artificial Intelligence
  • Knowledge Systems
Visit agentic-ocr-staging.nguyennhan-agenticocr.workers.dev All projects
Editorial illustration of a scanned document being transformed into structured data while a researcher reviews the source
Program artwork, not a screenshot of the software.

Research context

AgenticOCR sits at the intersection of Artificial Intelligence and Knowledge Systems. It asks how far a document-to-data workflow can be automated before a person has to check the source.

The program runs as a staging showcase: the workflow is live, and its status is stated rather than upgraded.

Approach

  1. Extract structured data from scanned documents, and keep the page it came from beside the result.
  2. Put review and editing inside the same workflow as extraction, so corrections stay attached to the source.
  3. Export in formats the next step can use: JSON, Excel, and HTML.

System architecture

Diagram — the AgenticOCR workflow. Read top to bottom: a scanned document is extracted into structured data, the source page stays beside the result for review, and the reviewed result is exported.

  1. Input

    Scanned document

    The source document, kept as the reference the extracted result is checked against.

  2. Extraction

    Document-to-data workflow

    Turns scanned pages into structured data.

    • Document-to-data workflow

  3. Review

    Page-level source review

    Keeps the source page beside the extracted result, so a reviewer can compare them.

    • Page-level source review

  4. Output

    Export

    Exports the reviewed result as JSON, Excel, or HTML.

    • JSON, Excel, HTML export

Current implementation

Published status
Staging showcase — live

The status above is the published one: the workflow can be tried, and the status stays staging rather than production.

What this does not claim

  • The program is not presented as a production system.
  • No accuracy or throughput figure for the extraction step is published here, so none is claimed.

Evidence

What the record establishes

The evidence is the live staging showcase and the workflow it exposes.

What it does not establish

A staging showcase is not a production release. The status is repeated here exactly as the site states it.

Roadmap

Every item below is planned, proposed, or under investigation. None of it is shipped, and this list is not a release schedule.

  1. Planned

    Move from staging showcase to a supported release

    Planned: a supported release, with the status updated only when it is true.

  2. Proposed

    Publish an accuracy report for the extraction step

    Proposed: a documented evaluation of extraction accuracy before any wider release.

Collaboration

AgenticOCR grows through document and research teams with real scanned collections and a review step that has to stay in the loop.

Partner with us How the lab works

Work on a program with us.

Each program takes on partners where the research question meets real practice — clinical review for AIMed, agent runtime governance for ActantOS and Agent Platform, academic workflows for ScienzaOS, and document-heavy collections for AgenticOCR.

Discuss a collaboration
  • Clinical pilots
  • Research collaborations
  • Dataset partnerships
  • Student theses
  • Open-source contributions