Kuken Kogyo's AI drawing search cuts lookups to an average of 4.5 minutes
A Japanese equipment maker built an OpenAI-powered search over decades of drawings, and projects a 90% cut in yearly labor spent on that task.
The short version
- Kuken Kogyo and Recursive built the system on the OpenAI API.
- It has two functions: drawing search and drawing check.
- In a factory trial, searches averaged 4.5 minutes.
- The firm projects a 90% cut in yearly drawing-search labor, based on fiscal 2025 results.
- Engineers still make the final call on every check.
Why it matters
Many companies have years of drawings, contracts or manuals that nobody can search well. This case shows AI helping with a bounded, checkable task instead of making decisions. Treat the headline figures as the company's own claim until you test the idea on your own files.
What the companies built
As monoist.itmedia.co.jp reported, Kuken Kogyo and the firm Recursive developed an AI-based knowledge system that runs on the OpenAI API. It is meant to help with equipment design and maintenance work.
The company has built up drawings and technical documents over more than 70 years. Digitizing them was already underway, but formats and notation vary, so finding the right document was slow.
Two tools in one web app
The first tool is drawing search. A user uploads a file such as a parts list, the AI pulls out the part numbers, and the system finds matching pages in the archive. It also surfaces near matches, such as differing model numbers, and shows a preview of each page so the user can verify it.
The second is drawing check. It compares the specifications in documents such as a work notice against those in an outline drawing or delivery spec sheet, then lists differences for a person to review.
The numbers, and their limits
In a trial at one of its plants, searches that used to take about an hour, or sometimes half a day, averaged 4.5 minutes. Using fiscal 2025 results, the company projects a 90% cut in annual labor hours for drawing search.
That 90% is a projection, not a measured result. The source does not say what the system costs or how the estimate was built.
People stay in charge
The system does not claim to catch every handwritten note or layout quirk. The AI's extraction and comparison are a starting point, and engineers make the final judgment.
Next, the company plans to improve how well it reads old drawings and handwriting, and to work on data upkeep, security and access controls. It aims for full rollout at two of its plants during 2026.
What to watch
- Does the 90% labor saving hold up once the system is in full use at two plants?
- Can it read old handwritten notes accurately enough to trust?
Try it
Time how long your team takes to find one past document today. Then pilot a search tool on a single archive and compare before and after on your own files.
Sources and links
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by monoist.itmedia.co.jp · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.