Home-HOOPS AI reaches general availability with new features to tackle the CAD-ML gap
Introduced as atechnology preview in late 2025,Tech Soft 3Dhas officially launchedHOOPS AI frameworkpurpose-built to integrate CAD data into machine learning (ML) pipelines.
The product is now generally available, building on a successful beta program with over 30 companies. The launch addresses a problem that has long complicated work for engineers in manufacturing and related fields. CAD datasets are notoriously difficult to feed into modern ML systems in any reliable way, and HOOPS AI takes that on directly by handling data preparation and model experimentation.
Speaking about the launch, Gavin Bridgeman, CTO, Tech Soft 3D said, “The official launch of HOOPS AI marks an important step in Tech Soft 3D leading the effort to bring AI to engineering data.”
New features, faster development cycles
The full release adds two capabilities that were absent from the preview. Linux support arrives alongside existing Windows compatibility, which is significant given that most ML infrastructure runs on Linux.
Alongside it comes CAD embeddings, a feature that automatically captures semantic relationships within CAD data without requiring human labeling. Rather than being told what to look for, the system identifies patterns on its own, allowing models to recognize similar parts and understand design context.
Teams can run hundreds or thousands of model variations simultaneously, which opens the door to tasks like part classification, metadata enrichment, manufacturing feature detection, similarity search, duplicate detection, and design reuse and optimization across large design libraries.
The ability to iterate at that scale is intended to compress development timelines, with Tech Soft 3D stating that smaller teams can cut cycles from months to weeks.
On the roadmap, Python access is set to expand with a specific focus on product manufacturing information (PMI). The company also intends to support training on private organizational data, a notable gap in the current version which has so far only been demonstrated on public datasets.
Source: 3D Printing Industry
