Texas-based DataStax has launched its Hyper-Converged Data Platform (HCDP) as a ready-made offering for enterprise customers that want to build vector databases for in-house generative AI (GenAI).
Vectors are mathematical representations of concepts, words and images used by GenAI systems to allow questions and comparisons of datasets to provide insightful outcomes.
HCDP is intended to allow companies to build GenAI platforms in their own datacentres so they can use the technology on their own private data. Integral to the release are Nvidia microservices and retrieval augmented generation (RAG) capability.
HCDP is not hyper-converged infrastructure in the sense of being a hardware appliance or combined storage and server software, but is intended by DataStax to be deployed cloud-like and virtual in the customer environment.
DataStax started out in 2010, and is rooted in the NoSQL database space. It also provides the cloud-based Astra DB database-as-a-service and DataStax Enterprise (DSE) for on-premise deployments. Version 6.9 of DSE comes alongside the launch of HCDP. Astra DB and DSE are both built on the Apache Cassandra NoSQL database.
HCDP is aimed at customers that want to build their own GenAI infrastructure on-premise, said Bill McLane, chief technology officer for cloud at DataStax.
“HCDP brings together generative AI and vector search capabilities to self-managed, enterprise data workloads,” he said. “So, companies that want to deliver generative AI services but don’t want to run in the cloud can achieve the same kinds of goals as cloud services can achieve.”
McLane said the product targets companies that want to run their own datacentres and ensure full control over how data is used in GenAI.
“With this in place, companies can use generative AI with their own data and large language models, but without handing that data over to a third party where they are not in control over how that data gets used in practice,” said McLane.
HCDP makes use of OpenSearch search and visualisation capabilities, as well as Apache Pulsar, a messaging platform used to build data pipelines, and load and distribute data.
Core to HCDP and DataStax’s functionality is the use of vectors in GenAI, said McLane. “Generative AI systems pull data together in response to potential questions based on vector search queries,” he explained. “These queries are turned into vectors – mathematical representations of the words or concepts involved – and then compared with the existing set of vector data that the company has. This information is then provided back to the large language model and used to prepare the response back to the user.”
Any data can be turned into vectors – product catalogues, previous customer histories or other unstructured data records – and can be stored ready for search.
Customers that want to use their own data in a GenAI system that uses a vector database alongside their transactional database can stream new data across to create and update vector data.
This can suit companies that want to keep control of their data for compliance and security reasons, or those that already have sizeable database installations and, for example, don’t want to migrate that data to the cloud for cost.