Overview
What this runbook does
This runbook describes, step by step, how an archive team processes a single batch of physical cultural assets (manuscripts, photographs, maps, periodicals, correspondence) into a fully catalogued, bilingual, quality-assured digital collection that is ready for publication. AI is used at three specific points in the pipeline: correcting raw OCR output, drafting a structured catalogue record for each item, and producing bilingual (Arabic and English) metadata. Every AI output is treated as a draft that a human must verify, and no record is published until a conservator and a cataloguer have both signed off.
The runbook exists because manual digitisation is slow and inconsistent, while fully automated digitisation is unsafe for irreplaceable heritage material. This process keeps humans in control of the decisions that matter (preservation, provenance, accuracy) while using AI to remove the repetitive drafting work.
Who this is for
- Digitisation operators who scan and register items.
- Cataloguers and metadata specialists who verify and enrich records.
- Conservators who assess physical condition and hold the publication veto.
- Digital collection managers who own the batch, the quality thresholds, and the publish decision.
This runbook is written for work carried out within the Department of Culture and Tourism, Abu Dhabi remit for culture, heritage, and the National Library. Substitute your own collection, storage, and cataloguing systems where placeholders such as <COLLECTION_NAME>, <DAMS_URL>, and <STORAGE_BUCKET> appear. Nothing here assumes a specific vendor product.
The outcome
When a batch completes this runbook you will have, for every eligible item:
- A preservation master scan and a derivative access copy in
<STORAGE_BUCKET>. - A verified full-text transcription (where the item contains text).
- A structured, standards-compliant catalogue record (Dublin Core plus local fields).
- Bilingual metadata (Arabic and English) with a recorded translation status.
- A documented conservation assessment and a signed conservator approval.
- A complete audit trail: who did what, which AI model produced which draft, and every quality-gate result.
Items that fail any gate are held, not published, and routed back to the responsible role with a reason. Nothing reaches the public collection by default.