slavon_ocr

Pipeline for OCR-ing scanned handwritten research cards — 20th-century scholar's index cards with excerpts from early 17th-century Ukrainian (Ruthenian) sources. Uses Claude Code with custom /ocr skill for transcription, preserving archaic Church Slavonic orthography.

Setup

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Usage

1. Extract pages from scanned PDFs

python pdf_to_jpeg.py scanned_pdfs/Auto-Color0002.pdf output/Auto-Color0002

2. Transcribe cards (requires Claude Code CLI)

Single card:

claude -p "/ocr @output/Auto-Color0002/001.jpeg"

Batch:

python batch_ocr.py output/Auto-Color0002
python batch_ocr.py output/Auto-Color0002 --force   # re-transcribe existing
python batch_ocr.py output/Auto-Color0002 --limit 10

3. Build HTML demo

python build_demo.py output/Auto-Color0002 -s tertiary
python build_demo.py output/Auto-Color0002 -s filename -o demo.html
python build_demo.py output/Auto-Color0002 --include-blank

Sort options: primary, secondary, tertiary, filename.

Blank/error cards are skipped by default; use --include-blank to keep them.

Output is a self-contained HTML file (images referenced via relative paths).