JevMade Sign in
← Back to experiments

Apps & data pipelines

tax-doc-classifier

Source screenshot of tax-doc-classifier
SOURCE SCREENSHOTFull screenshot ↗

What it does

Identify IRS forms and schedules from their text with Jev.

How you can use it

Start by collecting a few sample tax pages and listing the specific forms you expect to process. You can borrow the project's strategy: turn each page into text, send that text alongside form descriptions, and let an online service calculate the best match.

A developer can install the tool and set up an access key that connects your app to TypeSafe. Remember that scanned page text is transmitted to an external service, so you should ensure this fits your privacy rules for sensitive financial documents.

Maker-reported (not independently measured by JevMade): 0 wrong and 0 strict errors over 314 TaxCalcBench filled-form pages at $0.36 (author-reported, reproducible with pnpm eval) · 38 strict errors over 753 blank IRS form pages, 5.05%, at $0.86 (author-reported) · $0.00115 per page against $0.039 for the Sonnet classifier it replaces: 34x cheaper, ~0.5 s against ~3.3 s warm, and 261 forms named against 30 (author-reported comparison)

Primitives
choice
Platform
TypeScript
Added
Project created
GitHub stars
341 (snapshot, not live)

Keep this for later

Sign in to bookmark experiments, guides and videos, and keep notes only you can see.

Continue to sign in

We’ll bring you back to this listing.