BLUEPRINT Rebuilding a Legacy: Multimodal Retrieval for Complex Engineering Drawings and Documents
This AI finally cracks the code on decades of unsearchable technical archives.
Deep Dive
Researchers unveiled Blueprint, a multimodal AI system that automatically structures and searches massive archives of legacy engineering drawings. It processes ~770,000 unlabeled files by detecting drawing regions, applying specialized OCR, and fusing retrieval methods. On a benchmark of 5,000 files, it achieved a 10.1% absolute improvement in Success@3 and an 18.9% relative gain in nDCG@3 over leading vision-language baselines, making previously manual searches fully automated.
Why It Matters
It automates access to trillions in locked industrial knowledge, accelerating engineering and maintenance across critical infrastructure.