Classify a request, then review the result.

Run a classification workflow that gathers evidence, suggests a category when the evidence is clear, and holds the result for a person to review.

This standalone example uses synthetic support requests and phrase-matching rules. Classification and review run in your browser without a model or a Longhand account.

Run the example

Duplicate invoice charge

The customer sees two charges for the same invoice and asks which one is correct.

How the workflow works

  1. Scan the request description for phrases associated with billing, shipping, or account security.
  2. Suggest a category when the phrases point to one category. Leave the suggestion blank when signals conflict or evidence is missing.
  3. Have a reviewer confirm or assign a category. The reviewed result records the evidence, initial suggestion, and final decision.

Try it from the command line

Download both files to the same folder and run the sample with Node.js. The script prints evidence and proposals pending review. The browser example handles the review step.

node workflow.mjs sample-records.json

For a model-backed workflow, see how to plan private model training, inference, and approvals on hardware your team controls.

Read the Owned AI deployment guide