Digitizing the Variety print archive
An automated AI pipeline for Variety (Penske Media), consulting through Mission Cloud (2025): OCR and document understanding over a century of print issues, entity extraction, and persistence as structured data, for later use as a searchable archive.
Variety has covered the entertainment industry since 1905, and has been part of Penske Media since 2012. Its print run is well over a century of issues that exist as page images: readable by a person holding the right reel of microfilm, and effectively invisible to anyone searching. Consulting through Mission Cloud, I designed and implemented the first pipeline that started turning those pages into an archive you can actually query.
OCR is the easy half to describe and the hard half to do well. Old trade-press pages are dense multi-column layouts with mixed type sizes, advertisements wrapped around editorial, and print and scan quality that varies by decade. Running raw OCR across a page like that returns a soup of interleaved columns. So the pipeline does document understanding first — segmenting each page into articles, advertisements and listings — and only then reads each region in its own reading order, which is what makes the output a set of articles rather than a wall of text. On top of that sits entity extraction: the people, titles, companies and dates a piece is about, so an article becomes a set of things and not just a string.