################################
HOOPS AI 1.0-b2
################################

**Release 1.0-b2 (February 08, 2026)**


Technology Updates
==================


HOOPS Embeddings: Machine Learning Model for Shape Embeddings
-------------------------------------------------------------

This release introduces HOOPS AI's shape embeddings capability, enabling vector-based similarity search and retrieval for CAD parts using graph neural networks on B-Rep data.

See our :doc:`/programming_guide/shape-embeddings` page for more details.


New Features
============

**HOOPS AI 1.0-b2** comes with a set of brand new features such as :

Machine Learning & Retrieval:
-----------------------------

- **HOOPSEmbeddings**:

   - :doc:`/programming_guide/shape-embeddings` - Core embedding model for converting CAD files into high-dimensional vector representations
   
     * Built-in model registry for managing multiple embedding architectures
     * Automatic batch and parallel integration for efficient execution

- **FaissVectorStore**:

   - :doc:`/programming_guide/embeddings-retrieval` - High-performance vector storage and retrieval implementation
   
     * FAISS-based in-memory indexing for fast similarity search
     * Support for metadata filtering and CRUD operations
     * Persistent storage with save/load capabilities

- **CADSearch**:

   - :doc:`/programming_guide/embeddings-retrieval` - Unified interface for similarity-based CAD part retrieval
   
     * Shape-based search (query by CAD file)
     * Index persistence for efficient reuse across sessions

Key Capabilities:
-----------------

   - Generate embeddings for CAD parts using pre-trained models
   - Build searchable indexes from embedding batches
   - Query by shape similarity
   - Save and load indexes for fast startup
   - Parallel batch processing for large datasets

Other Improvements
------------------

- **GraphClassification** flow model now uses by default HOOPS Surface Encoder based on a BRep face mesh and a GAT architecture.
- **Flow pipeline** during data preprocessing will restart the process pool automatically to maintain a better memory print. Heavy files will be handle in sequential at the end.
- **Dataset Merging** is now twice faster. 
- **Dataset Loader** can now be instantiated with the list of graph files directly.

API Changes
===========

No APIs have been changed in **HOOPS AI 1.0-b2**

Documentation Changes
=====================

The HOOPS documentation has been updated to include new guides for Embeddings and Search.


Deprecations
============

No deprecations in **HOOPS AI 1.0-b2**

Fixed Bugs
==========

* Fix missing bar histogram using dataset Explorer in notebooks.