Industrial equipment manufacturer · Manufacturing & Industrial

Permission-aware engineering knowledge retrieval cuts documentation search time 42 %

A grounded retrieval system across drawings, standards, and twenty years of service history — respecting export-control and customer confidentiality permissions.

Challenge

Service engineers spent a third of their day searching for the right drawing revision, standard, or precedent across four document systems. Export-control and customer-confidentiality rules meant any search tool had to honour permissions exactly; a previous enterprise-search rollout had been withdrawn after a leak.

Approach

A Data & Platform Readiness review first mapped the sources and the access model. We then built ingestion with document-level ACL propagation, hybrid retrieval with reranking, and a grounded assistant that cites the exact page and revision. A golden set of 600 engineer questions drove evaluation throughout.

Solution

Hybrid search over 1.4 million documents with chunk-level citations, freshness guarantees via event-driven re-indexing from the PLM system, and refusal behaviour when confidence is low. A role-based enablement programme trained 300 engineers; a champions network now maintains the question set.

Results

−42 %

Time spent searching documentation

94 %

Answer faithfulness on golden set

0

Permission violations in 6 months

300

Engineers trained

“It answers with the drawing, the revision, and the page. Our engineers stopped arguing with it in week two.”
— VP Global Service

Find out where AI will pay off first.

A 30-minute discovery call, or the 5-minute readiness assessment. Either way you leave with a next step.