Kuken expects 90% search-time cut with OpenAI-based app

Kuken's archive problem is a familiar one for long-running manufacturers. Parts made today don't always fit older cooling towers, and without the original drawing engineers may have to design and manufacture a replacement from scratch. The drawings themselves record decisions made for particular installations, so the archive is as much a record of engineering judgment as it is a parts catalogue.

That context explains why search alone was not the goal. Kuken later digitized the archive, but engineers still had to know which name or number to look up and where the file was, and newer engineers spent time piecing those details together. Working with Recursive, Kuken tested several AI models on a range of drawings to see how accurately each could extract details and at what cost, and found OpenAI's models performed well on both measures.

The company is now trying to extend the same approach to review. In prototype testing of a tool that checks partner drawings against specifications, Kuken says about 90% of the issues the AI flagged were genuine discrepancies. If that tool moves into production, it would target a step engineers still do by eye today.