Historical records, made searchable

Genealogy runs on records: birth and marriage certificates, registers, and other official documents created over centuries by clerks who never imagined a search box. Most of them exist as images, handwritten or printed, in the language and layout of their place and time. Until the details on each record — names, dates, places, relationships — are pulled out and put into a consistent structure, the record can’t be found by the person searching for an ancestor.

Doing that by hand doesn’t scale. Researchers transcribing and verifying documents one at a time make slow progress through archives that hold millions of records, and fatigue brings errors. Fully automated reading has the opposite problem: it’s fast, but it stumbles on faded ink, unusual handwriting, and languages it wasn’t built for, and a single misread surname hides a record for good. The work needs machine speed and human judgment in the same process.

MLtwist combines them. A multilingual AI model pre-labels each record, drafting the fields so people correct rather than transcribe. Automated cleaning standardizes names, dates, and places so the same thing is written the same way everywhere. Reviewers verify the labeled data before it’s final, and the structured result is delivered as JSON that loads straight into a search database. For one leading genealogy company, that process made millions of historical records searchable.

It fits archives where the value is locked in document images and your users need to search them. The record images stay in your own cloud storage while the work runs, and nothing reaches your database until people have checked it. The same approach — AI extraction with human verification — also runs on modern documents, such as claims files for InsurTech.