Writing on archives, AI, and standards.

Untitled post

Why we built Archively.AI on standards, not promises

Every catalog tool claims to be 'ISAD(G) compliant'. Here's what that actually means in our data model — and what we refused to compromise.

AI drafts, curators decide: the case for three-layer state

How we separate AI output, curator review, and published snapshots — and why collapsing those layers is where most tools go wrong.

Transcribing oral histories: word-level confidence matters

The difference between 'good enough for search' and 'good enough to cite' is visible at the word level. A tour of our transcription pipeline.

EAD is dead. Long live EAD.

Thoughts on encoded archival description in 2026: still necessary, still ugly, still the only way to interop across national aggregators.

From spreadsheet to finding aid: a six-week pilot

How a specialist library moved from a CSV catalog to a published finding aid, without hiring a developer.

Vector search over archival prose: what actually worked

Our journey through pgvector, embedding choices, and the strange semantic texture of 19th-century correspondence.