Inventory Sheet OCR and Matching Pipeline

What it is

A staged Python pipeline that digitizes scanned warehouse inventory sheets (typed part descriptions with handwritten counts) for Chaparral Bolt & Supply. Claude Vision reads each page, descriptions are matched to the cleaned master parts list, and a confidence report plus local review apps let a human fix what the matcher was unsure about.

The problem

Thousands of handwritten inventory pages had to become structured counts keyed to the parts catalog.

How it works

1. Numbered scripts: fix part numbers, Claude Vision OCR per page keyed to bin location, assemble, match, confidence report, final match
2. Three-pass matching: exact normalized text, rapidfuzz token-sort fuzzy match, then Claude for the low-confidence remainder
3. Per-section runs with --dry-run and --limit flags, plus an alias file for known OCR misreads
4. Vite + Express review app and a separate CSV viewer for correcting roughly 7,900 OCR rows with save-back

Stack

Disciplines

For your business

Document digitization: turn scanned or handwritten paper records into structured, matched data with a human-review step.