5 advantages of using an integrated vector database for AI development

Vectors enable businesses to augment the LLMs used for generative AI. Here’s why the new vector data type should live in your existing database.

John Foley | January 15, 2024

As more businesses develop and deploy AI-driven applications, there’s a strategic decision to make: What vector database do we use?

Vectors, which are unique strings of numbers calculated to represent unstructured data, let companies add context to generic large language models (LLMs). Vectors enable rapid semantic search of the unstructured data they represent, a critical capability for uses such as making product recommendations or showing correlations among data or objects.

For many businesses, vectors are, or soon will be, an entirely new data type to manage. AI development teams must determine the best way to store, administer, and retrieve vector data. There are two primary options: specialized, purpose-built vector databases or multi-modal databases, such as Oracle Database 23ai, that support not only vectors but many other data types as well.

Searches on a combination of business and semantic data are easier, faster, and more precise if both types of data are managed by a single database.”

Juan Loaiza Executive Vice President, Mission-Critical Database Technologies, Oracle

There are benefits to both approaches. Specialized vector databases serve the purpose of letting an LLM use your data when responding to queries. However, they may not be as well suited to other data types and workloads. And because vector databases are new, they must be integrated into your existing application architecture. That work includes figuring out scalability, adding security and identity management, and meeting availability and performance expectations.

All-purpose databases like Oracle Database 23ai avoid those issues. Not only does Oracle Database 23ai handle many data types, including vectors, but it’s integrated into your application environment and already contains your company’s data. There’s no need to move data into a specialized vector database. Your team can focus its efforts on augmenting an LLM with company data.

The term Oracle uses to describe this kind of highly integrated model is a “converged database,” that is, a database with native support for all modern data types, analytics, and the latest development paradigms. Oracle Database, for example, supports transactions, analytics, AI/machine learning, blockchain, graph, spatial, JSON, REST, events, IoT streaming, and more—all as part of the core system.

“It allows you to support many diverse projects using a single platform,” wrote Maria Colgan, distinguished product manager at Oracle, in a blog post on converged databases.

Oracle recently added vector data to the growing list of data types incorporated into Oracle Database. This support comes in the form of a new, limited availability1 capability in Oracle Database 23ai called “AI Vector Search.” It includes vectors as a native data type, as well as vector indexes and vector search SQL operators, which together make it possible to store the semantic content of unstructured data as vectors. You can then run split-second similarity queries on documents, images, and any other unstructured data represented as vectors.

Easier, faster, more precise

In announcing AI Vector Search at Oracle CloudWorld in September, Juan Loaiza, Oracle executive vice president of mission-critical database technologies, emphasized the discoverability advantages of integrating vector and business data.

“Searches on a combination of business and semantic data are easier, faster, and more precise if both types of data are managed by a single database,” Loaiza said.

You can begin to see the business case for using a converged database to store and retrieve vector data. And there are other compelling reasons to consider this approach, which I summarize below.

Five advantages of using an integrated vector database:

1. Versatility. Converged databases handle a wide range of data types and workloads. They accommodate not only the vector-enabled applications that are a growing priority for many companies, but their built-in flexibility leaves the door open to new use cases involving other data types as they emerge. Converged databases aren’t one-trick ponies.

2. Less complexity. For years, IT leaders have grappled with database sprawl, the result of departmental projects, specialized databases, point solutions, and “shadow IT” making its way into data infrastructure. The last thing CIOs and CTOs want is another one-off platform. Oracle Database 23ai helps ease complexity by serving as a corporate standard for a wide range of data-management requirements—transactions, analytics, AI, geographic distribution, data consolidation, and more.

3. Combined structured and unstructured data. With AI Vector Search, Oracle Database 23ai can blend structured business data with unstructured vector data, a capability that Loaiza demonstrated in a prototype house-hunting application at Oracle CloudWorld. An added benefit of this integrated approach is that it reduces the need to move or synchronize data across databases, enhancing consistency.

4. Existing skills. Does your organization have the expertise and hands-on administrative resources necessary to set up and manage a specialized vector database? If not, another advantage of using Oracle Database 23ai for vector search is that many developers and DBAs already have experience with Oracle Database.

5. Enterprise-class capabilities. As vector-enabled applications advance from pilot projects to customer-facing deployments, they must deliver the levels of performance, scalability, security, and reliability that business managers expect from run-the-business applications. Oracle AI Vector Search clears that bar by leveraging other enterprise-class Oracle capabilities, such as Real Application Clusters (RAC), partitioning, sharding, security, analytics, and disaster recovery.

More AI building blocks

As these examples show, Oracle Database 23ai can be an excellent way to add vector-enabled similarity search to the user experience. And AI Vector Search is just one of several new AI building blocks available from Oracle.

For example, applications built on Oracle Database and Autonomous Database can add an LLM-based natural language interface. In fact, Oracle Autonomous Database released a natural language interface, called Select AI, in September 2023. And Oracle Database tools APEX and SQL Developer offer generative AI capabilities, currently in limited availability, that let developers use natural language to generate applications or SQL queries.

The 23ai database is built to empower developers as well as data professionals. As technology decision makers evaluate their options for building AI applications that can combine internal vector data with LLMs and provide natural language interfaces, they should look closely at the new capabilities in Oracle Database 23ai.

1 The AI Vector Search capability is available today in the Oracle Database 23ai free edition for a select group, with broader access coming in April 2024.

John Foley is Editor of the Cloud Database Report and a vice president with Method Communications.

This article was updated on May 8 to account for an Oracle Database 23c name change to Oracle Database 23ai.

View more Oracle Connect articles