Computer Vision

OCR — Image to Text

Read text from photos and scans, in-browser.

runs 100% in your browser — nothing is uploaded

Image with text (photo, scan, screenshot)

How it’s engineered

Tesseract compiled to WebAssembly: ~5 MB fetched once, then OCR runs locally; documents never leave your device.
Feeds the Invoice → JSON demo: camera photo to accounting-ready record without a single server.
In production pipelines I pair OCR with layout detection and an LLM cleanup pass for handwriting and low-quality scans.