Hazelcast, Inc., a leading software provider powering mission-critical applications that move the economy, is debuting vector search capabilities for its flagship product, the Hazelcast Platform. Now providing the necessary tools for Hazelcast users to modernize for the AI era, Hazelcast Platform 5.5 enables a variety of new use cases—including semantic search, fraud detection, and retrieval augmented generation (RAG)—as well as additional advancements in compute, resilience, and continuity.
Support for vector search empowers organizations to deploy high-performance, scalable pipelines to query structured and unstructured data. The Hazelcast Platform now delivers the necessary agility to produce vector data structures and embeddings from text plot summaries, opening up a new world of data scientist efficiency, according to the company.
“The ability to have that context—have that reference data alongside the actual vectorized images, which you can then search for—is really the benefit of having something like Hazelcast at your core,” said Avtar Raikmo, VP of engineering at Hazelcast. “Being able to mine that data and target data science use cases is really critical. Data scientists typically don't have the expertise of being able to build industrialized software, which is easy to support and scale, is performant, is resilient, and all of the other benefits that a platform like Hazelcast can bring.”
Additionally, the Hazelcast Platform has demonstrated significant performance gains over most of its competitors, particularly in the space of vector embeddings and retrievals. When benchmarked against 1 million OpenAI angular vectors, Hazelcast outperformed by delivering single-digit millisecond latency when uploading, indexing, and searching vectors with 98% precision, according to the company.
“Many vendors have highly optimized algorithms for data retrieval, but you pay the penalty on vector creation and insertion,” noted Raikmo. “The algorithms that we've ended up going with—which are based on disk-based approximate nearest neighbor—really allows us to have much more of a 50/50 distribution between computation benefits as well as storage and retrieval benefits.”
In conjunction with support for vector search, Hazelcast Platform 5.5 delivers two new capabilities that afford organizations with greater agility, resilience, and performance for enterprise applications:
- Jet Job Placement Control: Customers can separate compute of Hazelcast Platform nodes from data store components to drive flexibility and resilience for compute-intensive workloads
- Client Multi-Member Routing: Enhances the resilience, throughput, and control of applications that are connecting to geographically scattered clusters
Specifically, in regard to Jet Job Placement, Raikmo explained that “in the world of AI, you want to have some members within a cluster that are more efficient at doing computation and others that are more efficient at storing the data and making it available for retrieval. So, with the ability to actually now disaggregate your compute and storage needs, vector search is really going to be the main beneficiary of that.”
Hazelcast Platform 5.5 also introduces an industry-leading 3-year long-term support (LTS), ensuring mission-critical customers can build for the long term with streamlined upgrades, according to the company.
To learn more about Hazelcast’s latest update, please visit https://hazelcast.com/.