WorkVellum Archive
Vellum
- Challenge
- A catalogue you had to already understand
- Outcome
- Full-text search in under 200ms
- Client
- Vellum Archive
- Services
- Data & Integration + Web & Mobile
- Industry
- Public Sector & Archives
- Year
- 2023
vellum
Vellum search results across the archive
The challenge
1.4 million scanned documents behind a catalogue you had to already understand. If you knew the reference number you were fine. If you were a researcher with a question, you were stuck.
Full-text search existed and returned 40,000 results ranked by scan date, which is the same as returning nothing.
vellum-vellum-
Our approach
We rebuilt the index around what researchers search by rather than what the scanning process recorded — place, period, document type, named people — and accepted fuzzy matching on OCR text that is often wrong in predictable ways.
The viewer matters as much as the search. Finding a document you cannot read is not finding it.
- Documents indexed
- 1.4M
- p95 search latency
- <200ms
- Sessions reaching a document
- 3.4x
- Engagement
- Discovery → Project, 12 weeks
- Stack
- Python, OpenSearch, PostgreSQL, React
- Accessibility
- WCAG 2.2 AA, screen-reader tested
Gallery
vellum-g0
vellum-g1