Kuken Finds Drawings in 4.5 Minutes

Kuken Industries, a Japanese maker of air conditioning and ventilation equipment, built an AI search tool with Japanese startup Recursive using the OpenAI API. Engineers can find a drawing for a replacement part in 4.5 minutes, down from nearly an hour, and Kuken projects a 90% reduction in annual hours spent searching. The company holds one of Japan’s largest shares of the cooling tower market.

Kuken’s archive covers 70 years of installations. Equipment was adapted to individual customers, so specifications vary by customer and installation date; parts made today may not fit older towers. Before digitization, a search could take half a day. Even after records were digitized, engineers needed to know which name or number to search and where to find the file, leaving newer staff reliant on experienced colleagues. Kuken wanted engineers at every experience level to find records independently and learn from the decisions recorded in them.

Recursive and Kuken built the search system and a field app using the company’s archive and engineers’ input. They tested several AI models on drawings for extraction accuracy and cost. OpenAI’s models performed well on both measures, and the OpenAI API allowed search to be built directly into the app. Engineers can enter a drawing or part number, or upload an image or PDF of a drawing or parts list. The app reads numbers from screenshots and returns possible matches, including drawings for similar designs; engineers check tower specifications and past replacement records to identify the correct drawing.

Kuken says experienced engineers use some saved time to mentor newer colleagues, while in-app drawing reviews reduce printing costs. Newer engineers now handle a wider range of jobs independently, and staff say they have more time to study drawings, consider design improvements and explore ideas. Experienced colleagues also have more time to teach and do engineering work themselves.

Kuken is developing a separate tool to review drawings prepared by partners. Engineers currently compare each drawing with specifications by eye. In prototype testing, about 90% of the issues flagged by the AI were genuine discrepancies, Kuken says. If the tool enters production, the company expects it to substantially reduce checking time. Kuken says making past installation decisions easier to find and understand can help pass engineering experience to the next generation.

Kuken Industries, a Japanese maker of air conditioning and ventilation equipment, built an AI search tool with Japanese startup Recursive using the OpenAI API. Engineers can find a drawing for a replacement part in 4.5 minutes, down from nearly an hour, and Kuken projects a 90% reduction in annual hours spent searching. The company holds one of Japan’s largest shares of the cooling tower market.

Kuken’s archive covers 70 years of installations. Equipment was adapted to individual customers, so specifications vary by customer and installation date; parts made today may not fit older towers. Before digitization, a search could take half a day. Even after records were digitized, engineers needed to know which name or number to search and where to find the file, leaving newer staff reliant on experienced colleagues. Kuken wanted engineers at every experience level to find records independently and learn from the decisions recorded in them.

Recursive and Kuken built the search system and a field app using the company’s archive and engineers’ input. They tested several AI models on drawings for extraction accuracy and cost. OpenAI’s models performed well on both measures, and the OpenAI API allowed search to be built directly into the app. Engineers can enter a drawing or part number, or upload an image or PDF of a drawing or parts list. The app reads numbers from screenshots and returns possible matches, including drawings for similar designs; engineers check tower specifications and past replacement records to identify the correct drawing.

Kuken says experienced engineers use some saved time to mentor newer colleagues, while in-app drawing reviews reduce printing costs. Newer engineers now handle a wider range of jobs independently, and staff say they have more time to study drawings, consider design improvements and explore ideas. Experienced colleagues also have more time to teach and do engineering work themselves.

Kuken is developing a separate tool to review drawings prepared by partners. Engineers currently compare each drawing with specifications by eye. In prototype testing, about 90% of the issues flagged by the AI were genuine discrepancies, Kuken says. If the tool enters production, the company expects it to substantially reduce checking time. Kuken says making past installation decisions easier to find and understand can help pass engineering experience to the next generation.