Document Vision

Converting paper medical records to digital is difficult at baseline, but a couple of factors make it extra challenging in Ethiopia: documents are often in Amharic (or a mix of Amharic and English), and access to the compute resources required to run state of the art OCR models is limited. Tesseract is fits this situation nicely in my opinion; It can run cheaply on small on premises servers, and has support for Amharic out of the box.

Document Vision is a demo app for learning how to configure Tesseract for best performance on Amharic documents. It lets you upload a document, run OCR with a preset, inspect each preprocessing step, review confidence scores, and compare recognized text against detected word boxes.

The preset controls make practical OCR tradeoffs visible: grayscale conversion, rotation detection, deskewing, resizing, binarization, noise reduction, and Tesseract page segmentation settings.