ServicesBlogPricingContactContact Now
AI4 min read

Kuken Industries Cuts Drawing Search Time 90% With OpenAI API

Kuken Industries Cuts Drawing Search Time 90% With OpenAI API

Introduction

Japanese cooling tower manufacturer Kuken Industries has deployed a generative AI search system built on the OpenAI API that reduces technical drawing retrieval time from nearly an hour to 4.5 minutes. The tool unlocks 70 years of installation-specific engineering decisions, allowing less-experienced engineers to locate and interpret legacy designs without relying on senior colleagues for guidance.

The Challenge: Legacy Knowledge Management in Industrial Engineering

Kuken holds a leading share of Japan's cooling tower market, supplying equipment for commercial buildings, factories, and data centers. Over seven decades, the company accumulated a vast archive of installation drawings, each reflecting custom specifications for individual customers. When a replacement part is needed, engineers must identify the exact drawing for that tower's original configuration, because parts manufactured today often do not fit older units without modification.

Even after digitizing the archive, search remained cumbersome. Engineers needed to know precise model numbers, part names, and file locations. Newer staff frequently spent hours piecing together those details or waiting for experienced colleagues to point them in the right direction. The knowledge gap created a bottleneck: senior engineers were consumed by search assistance rather than higher-value design work, while junior engineers could not develop the pattern-recognition skills that come from studying historical decisions.

The Solution: Multimodal AI Search Architecture

Kuken partnered with Recursive, a Japanese AI startup and official OpenAI partner, to build a search application that accepts multiple input modalities. Engineers can enter a drawing number, part number, or upload an image or PDF of a drawing or parts list. A screenshot of a parts list is automatically parsed for part numbers, returning candidate matches that include drawings for similar designs. The engineer then verifies the correct match against tower specifications and past replacement records.

The development team evaluated several AI models on extraction accuracy and cost across Kuken's heterogeneous drawing set. OpenAI's models performed best on both dimensions, and the API's flexibility allowed the search capability to be embedded directly into a field-ready application. Continuous feedback from field engineers shaped the interaction model, ensuring the tool fit actual workflows rather than a theoretical ideal.

Measurable Impact: Quantifying Efficiency Gains

Kuken projects a 90% reduction in annual hours spent searching for drawings. The time savings cascade across the organization:

  • Junior engineers now handle a broader range of repair jobs independently, accelerating their development.
  • Senior engineers reclaim mentoring capacity and can focus on complex engineering tasks.
  • Drawing review shifts from paper to the app, cutting printing costs and encouraging deeper study of design rationale.
  • Design improvement becomes feasible because engineers spend less time hunting for information and more time analyzing it.

Both veteran and newer engineers report that the reclaimed time creates space for considering design improvements and exploring ideas that were previously deprioritized due to search overhead.

Extending AI to Drawing Review: Quality Assurance Automation

Building on the search success, Kuken is prototyping a drawing verification tool for partner-submitted designs. Today, engineers manually compare each drawing against specifications, checking for mismatched details or quantities. The AI tool flags discrepancies automatically. In prototype testing, approximately 90% of flagged issues were genuine discrepancies. If promoted to production, Kuken expects a substantial reduction in drawing review time and a lower miss rate for human reviewers.

Practical Takeaways for Engineering Organizations

Organizations with deep technical archives should consider the following steps:

  • Audit search friction. Measure how long subject-matter experts spend locating legacy artifacts versus applying them.
  • Prioritize multimodal input. Field conditions often favor photo or screenshot upload over manual data entry.
  • Validate model choice on real data. Accuracy and cost trade-offs vary significantly across document types; test on a representative sample before committing.
  • Involve end users in UX design. The Kuken team shaped the app around engineer workflows, not the other way around.
  • Plan for knowledge transfer. The highest ROI may come from enabling junior staff to learn from historical decisions, not just from raw time savings.
  • Extend to adjacent workflows. Once search is solved, review, verification, and change-management processes become natural next targets.

Grivyonx Expert Analysis

Kuken's case illustrates a pattern we see across industrial sectors: the highest-value AI deployments are not chatbots but domain-specific retrieval and verification systems that sit on top of proprietary, unstructured data. The 90% search-time reduction is impressive, but the strategic win is decoupling junior engineer productivity from senior engineer availability. For security and compliance teams, note that the architecture keeps drawings on-premises or in controlled cloud tenancy while only sending query embeddings or cropped regions to the model API, a pattern that satisfies data sovereignty requirements common in critical infrastructure supply chains.

Conclusion

Kuken Industries demonstrates that generative AI can unlock decades of trapped engineering knowledge without requiring a full data restructuring project. By combining multimodal search with a field-first user experience, the company turned a 70-year archive from a liability into a live mentoring tool. As more industrial firms face workforce transitions and legacy equipment fleets, this retrieval-first approach offers a repeatable blueprint. Grivyonx helps organizations implement similar AI-powered knowledge extraction pipelines within secure DevSecOps frameworks tailored for operational technology environments.

  • #AI in manufacturing
  • #OpenAI API
  • #knowledge management
  • #industrial engineering
  • #multimodal search
  • #legacy systems modernization
Gourav Rajput

Gourav Rajput

Founder of Grivyonx Technologies