A replacement part search that once took nearly an hour now takes 4.5 minutes at Kuken Industries. The Japanese industrial manufacturer, which holds a significant share of Japan’s cooling tower market, deployed an application built on the OpenAI API to search through seven decades of technical documentation. For industrial operations burdened by decades of customized equipment, integrating machine learning models into archive management turns unstructured legacy records into searchable assets, cutting labor overhead and unblocking internal maintenance cycles.
Kuken Industries produces air conditioning and ventilation equipment for large buildings, shopping centers, and factories. Its cooling installations span a 70-year operational history, with equipment adapted over time to individual customer requirements and specific installation dates. Because parts manufactured today frequently fit older cooling towers poorly, maintenance teams require original technical drawings. Lacking those specifications, engineers often faced the cost of designing and fabricating replacement components from scratch.
Overcoming the Digitization Bottleneck
Standard archive retrieval previously created substantial operational drag. During the paper-archive era, locating a specific blueprint could consume half a day. A subsequent transition to digital storage removed the physical file cabinets but preserved the underlying friction: engineers still had to know specific file locations and exact part designations. When standard model or part queries returned multiple conflicting results, newer staff had to rely directly on veteran personnel to locate records and interpret historic design decisions.
To resolve this knowledge bottleneck, Kuken Industries partnered with Recursive, a Japanese AI startup and official OpenAI partner. The two companies jointly designed a search architecture and dedicated field application informed by Kuken’s physical archives and direct input from its maintenance workforce.
"I believe OpenAI’s technology can help manufacturers like us pass skills on to the next generation. That’s long been a challenge for us."
As recorded by Kuken, the imperative was enabling newer staff to independently retrieve records, analyze past design decisions, and execute repairs without pulling senior personnel from core engineering assignments.
Pipeline Architecture and Measured Returns
During development, Recursive and Kuken Industries evaluated multiple AI models across sample drawings, testing extraction accuracy against compute expenditure. OpenAI’s models delivered the required extraction fidelity within acceptable cost boundaries, while the OpenAI API allowed direct integration into the operational field interface. The resulting workflow allows technicians on site to enter part numbers or simply upload image files and PDF documents of component lists and technical diagrams. The software processes screenshots, cross-references historical installations, and returns potential matches alongside related equipment specifications.
"People now have time to do work they’ve wanted to try and take on new challenges. That’s been a real source of motivation."
As Kuken reported, the deployment altered team capacity and operating expenses. Kuken Industries projects the application will cut drawing search hours by approximately 90% annually. Technicians review files directly inside the software interface, lowering corporate printing overhead. Freed from manual search guidance, veteran engineers reallocate saved hours toward junior mentorship and specialized development tasks, while newer staff resolve legacy repairs autonomously.
Kuken Industries is also developing an automated drawing-checking tool to audit blueprints submitted by external partners. In prototype testing, approximately 90% of the discrepancies flagged by the system represented genuine errors, providing a viable blueprint for eliminating manual visual audits in industrial manufacturing.
What’s behind it
Does deploying large AI models deliver real value in heavy manufacturing burdened by seventy years of custom archives? The immediate payoff often lies in removing friction for field technicians rather than automating core engineering from scratch. By enabling junior staff to pull historical schematics directly without distracting senior designers, an enterprise reclaims high-value engineering hours while eliminating the costly remanufacturing of legacy parts. According to Kuken Industries, the API integration cut parts retrieval time from nearly an hour down to 4.5 minutes.