Preparing documents for ingestion
What to load first, what the pipeline handles, and how to keep the corpus healthy.
What to load first
Start with the documents people actually ask about: current policies, procedures, playbooks, precedents. A small corpus of live material beats a huge archive of maybes — you can always widen scope once the first team trusts the answers.
What the pipeline handles
Real corpora are messy, and the pipeline was built for that:
- Scanned PDFs and image-heavy files, including poor-quality scans
- Long documents, appendices and mixed formats
- Spreadsheets and structured exports
- Legacy files that predate whoever now owns them
If a document defeats ingestion, it surfaces as such rather than silently vanishing — ask your administrator to review the media library for anything flagged.
Corpus hygiene
- Prefer current versions. Where possible, scope live policy sets rather than archives, so answers come from what’s in force.
- Use meaningful locations. Connectors ingest what you scope; tidy source folders make scoping precise.
- Let the knowledge network help. As documents arrive, Marella extracts entities and concepts — reviewing them weekly keeps the ontology clean while the corpus is young.
When in doubt
Bring the awkward cases to your OpenKit contact. The unusual documents are usually the most valuable ones to get answering.