JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Domain implementation guide

Classify tax-document pages with bounded form criteria

A domain-specific classifier that extracts PDF page lines and asks Jev to choose among 261 IRS forms and seven page kinds using a versioned criteria file derived from IRS PDFs.

Original by kyotofinClassificationGitHub README and classifierSource reviewed

Before you dive in

What you’ll find in the original

  1. Install Poppler so PDF pages can be converted into the line input expected by the classifier.
  2. Keep form knowledge in the criteria JSON and pass it with page lines to `classifyPage` instead of training or hosting a custom model.
  3. Retain the returned probability distribution and evaluate uncertain pages rather than reducing every result to an unchecked top label.
Worth knowing

Requires Poppler, Node.js, and TypeSafe credentials. The project's 34× cost and 6× speed comparisons refer to its prior internal pipeline and are not independently reproduced or universal.

More guides →See the experiments →