AI Cataloging for Archives — Photos, Files, Films
The backlog is not a scanning problem. It is a description problem. Photographs, files and films sit undescribed, so researchers cannot find them. This engine drafts a full catalog record for each item, in the archive's working languages. A cataloger reviews, corrects and signs off, field by field.
Municipal, national, university, corporate, religious, community and research archives are in scope. The engine learns that archive's fields, vocabularies and languages. It does not ship with a domain thesaurus that every client has to wear.
This is content description, not another OCR box. Transkribus reads text. ArchivesSpace and AtoM store records. We fill the record: title, dating, places, persons, historical context, a name index, and an explicit confidence mark on every identification.
What goes in
- Photographs — including the verso: archive numbers, dates, handwritten notes and dedications, across languages.
- Files / dossiers — either a concise file-level description (one record for the whole file) or page-level transcription.
- Films — scene timeline, intertitles, bilingual records.
Mixed-language, mixed-script collections are the normal case, including RTL and handwriting.
What comes out
A genuine catalog record, not a keyword dump:
- Fields written as cataloging prose in the languages the archive works in — not a machine-translation dump after the fact.
- Confidence markers on identifications: confirmed / probable / tentative. The record does not sound sure where it is not. Enforced in code, not as a slogan.
- Never invent. If the source does not say it, the field stays empty.
- Export into the system you already use: EAD, Dublin Core, MARC21, and institutional ingest sheets. We do not replace ArchivesSpace, AtoM or your CMS. We feed them.
How it is run
A dedicated server per archive. No shared tenancy. Client data is not used to train models. Keys stay under the archive's control. For material that must not leave the building, processing can stay local.
The cataloger is not optional. Delicate identifications — who a person is, which place, which event — remain a human decision. The engine proposes and rates itself; only a person promotes a draft to a record.
A free pilot, on your material
Organizations can run a free pilot of 100 documents, 10 photographs and 5 films, with no time limit; individuals can catalog 10 items free. Include the hard items. You compare the draft records with what you would have written, then decide. That is the sales motion. Not a slide deck.
Request the pilot on the cataloging page, or write to [email protected].
Related: the cataloging engine in more detail, audit trails and citations.