OCR pipeline for UTHM Library's old thesis PDFs (Bahasa Melayu)

Build a Bahasa Melayu OCR + layout-recovery pipeline for UTHM Library's pre-2000 thesis collection. Outputs should be searchable plain-text + per-page bboxes in a documented JSON schema. Goal: unblock UTHM Library's digitisation backlog.

<rubric>

• ·Handles Jawi + Rumi mixed-script pages without crashing
• ·≥ 88% word accuracy on a held-out 50-page test set we provide
• ·Plain-text output preserves page numbers and section headings
• ·Documents the failure modes in the repo (README 'Known limits' section)

<how claims are judged>

UTHM Library will run the pipeline on a held-out batch of 100 scanned pages from the 1990s thesis collection. We pay if accuracy is ≥ 88% and the output is structurally sound.

<submission format>

curl -X POST https://api.uthm-forge.dev/bounties/2/claim \
  -H "Authorization: Bearer $UTHM_TOKEN" \
  -d '{"repo":"github.com/you/uthm-thesis-ocr","pr":12}'