DataStax today seeks to make it easier for developers to build generative AI retrieval augmented generation (RAG) applications with a new data API.
DataStax is one of the leading commercial vendors behind the open source Apache Cassandra database, which is the foundation of its AstraDB cloud database as a service. Like many other database vendors, DataStax added vector database capabilities to its platform in 2023. At a recent event, the CEO of DataStax claimed that Cassandra is “.. the best f*cking database for gen AI.”
The vector database capability is critical to enabling RAG applications that combine large language models (LLMs) with data platforms to produce highly accurate and customizable results.
While DataStax has vector capabilities in AstraDB since July 2023, this capability still requires users to work with Cassandra Query Language (CQL) as the primary way to query data. Today’s new Data API changes that, giving developers the ability to use the Python and JavaScript programming languages to access the database, which the company claims is DataStax and purpose-built vectors. Helps bridge the gap between databases like Pinecone, which just updated its name. Platform with serverless database functionality.
“There’s kind of a war going on between messy vector databases that don’t support any query type other than vectors and hybrid databases,” Ed Enough, chief product officer at Datastacks, told VentureBeat. “We had to close that gap and that’s what the History API is all about.”
How the DataStax data API changes the way developers build RAG applications.
The new data API does not provide any new vector capabilities to the AstraDB database. What it does instead is make it easier for developers to build applications.
According to Inf, the goal of the new API is to reduce the bottleneck mismatch between what developers do and what the database provides. Info noted that since July 2023, when Vector’s capabilities first landed in AstraDB, nearly half of all new customers signing up for the cloud database are using it to build next-gen AI applications.
The challenge is that those developers weren’t able to easily use the programming languages they were already using to build gen AI applications, which is mostly Python and JavaScript, to access AstraDB.
Before the new data API, developers building AI applications with AstraDB had to use the standard Cassandra Query Language (CQL), which requires more data modeling knowledge than developers want to deal with for simple rack applications. were Queries also don’t perform so well for vector data.
Inf explained that the new data API improves performance by automatically handling vectorization, offering a simple interface to languages like Python and JavaScript, and storing and indexing vector data more efficiently at the database level. Instead, add vectors as another data type. . This reduces the learning curve and improves performance compared to building on top of existing Cassandra APIs and data models.
It’s all about APIs.
With some classes of database APIs, all that happens is a form of translation from a native programming language, such as Python or JavaScript, to whatever query language is used for the database. This is very similar in practice to the decades-old approach of how developers worked with databases through an Object-Relational Mapper (ORM).
The DataStax data API is a little different because Cassandra is built differently from other databases. At the architectural level, Cassandra is organized around a set of high-performance primitives that combine together to support a variety of query patterns. The Cassandra data architecture makes it possible to connect to a deep layer in the database, improving overall query performance, Inf said.
“The Data API exposes a very simple JSON-based data format to the developer, where you can describe anything within JSON, the developer can send to and retrieve from the database,” said Inf. said “But we store it in a very efficient way inside Cassandra where we do it directly on the storage tier and make sure that the performance that a developer gets is maintained.”
Vector acceleration with JVectorEngine
Another important part of DataStax’s vector database development is the JVector search engine that is part of AstraDB. JVector is an open source embedded vector search engine developed by DataStax.
Inf explained that JVector uses an algorithm called DiskANN, which is a disk-based storage-optimized version of the ANN (Approximate Nearest Neighbor Search) algorithm that is widely used in almost all vector databases. They noted that DiskANN provides significantly better retrieval capabilities than other algorithms that do not perform well at large storage and distribution scales.
According to DataStax, the JVector engine is what allows AstraDB to have better compatibility and recall than other vector databases. Much of DataStax’s vector functionality, including JVector and the data API, is being open sourced for use by the Cassandra open source community as well as DataStax’s AstraDB customers.
“We’re committed to making things available to open source ecosystems,” said Inf. “We just want to make sure that if you’re just a developer trying to figure out what you need Should be using a cloud service, that you’ve found the easiest way to do it.”
VentureBeat’s Mission To be a digital town square for technology decision makers to learn about transformative enterprise technology and transactions. Explore our briefing.