OCR Pipeline

The problem

A campus mailroom serving 70,000+ students was hand-keying every package into a housing portal with no API — 1,000 to 1,500 labels a day of pure manual entry that slowed pickups and burned staff hours. The insight: the bottleneck wasn't the volume — it was a closed system nobody could automate against.

What I built

• Google Cloud Vision OCR reads student names, tracking numbers, and carrier info straight off the label photos.
• Fuzzy matching handles the messy middle — reconciling nickname-vs-legal-name mismatches like “Abby” and “Abigail” so the right student gets tagged.
• Python + Selenium drive the API-less eLiving portal the way a person would, auto-filling each entry — making a closed system programmable instead of waiting on one that would never open up.
• Auto-triggered email notifications close the loop, so students hear about a package the moment it's logged instead of after a manual sweep.

Impact

• Runs in production, processing 1,000–1,500 packages daily across a 70,000+ student population — replacing manual data entry and cutting pickup delays so staff supervise rather than type.
• Fully FERPA compliant by design: no PII stored or exposed outside secure internal systems — the compliance constraint shaped the architecture from the start, not as an afterthought.
• I built and own the pipeline end-to-end — OCR, the fuzzy-matching layer, the browser-automation bridge into a system with no API, and the notification loop.