Introducing AlloyDB AI for PostgreSQL

Not just for chatbots that respond in a human-like manner, but also for the way it can open up completely new user experiences, generative AI has captured our attention in countless ways. Additionally, these new generation AI workloads are accessible to a wider range of the developer community than traditional AI workloads, which call for additional specialized skills. The innovation in next-generation artificial intelligence applications will lie not just in the models themselves but also in their application and the data that underpins it.

At Google Cloud Next today, we introduced AlloyDB AI, an integrated set of features built into AlloyDB for PostgreSQL that will assist developers in creating scalable and effective generation AI applications using operational data. By offering built-in, end-to-end support for vector embeddings, AlloyDB AI enables developers to more quickly and effectively combine the power of large language models (LLMs) with their real-time operational data.

With a straightforward SQL function for creating in-database embeddings, AlloyDB AI enables users to quickly transform their data into vector embeddings and executes vector queries up to ten times faster than standard PostgreSQL. An end-to-end solution for developing next-generation AI applications is offered by integrations with the open source AI ecosystem and Google Cloud’s Vertex AI platform.

Later this year, the AlloyDB managed service will introduce AlloyDB AI, which is currently available in preview via the downloadable AlloyDB Omni.

Using vector embeddings, you can use LLMs and your data together

Enterprise generation AI apps must overcome a number of obstacles that LLMs cannot solve on their own. These apps must deliver accurate and current information, offer contextual user experiences, and be simple to create and manage for developers.

The link between LLMs and enterprise generation AI apps is provided by databases with vector support. Why? The most recent information for your users and applications is in databases, first. With techniques like Retrieval Augmented Generation (RAG), you can ground LLMs in real-time data from your database, enhancing accuracy and assisting in making sure that responses are instructive, pertinent, actionable, and individualized for the user.

You can retrieve information based on semantic relevance thanks to the support for vector embeddings, which are numerical representations of data used to represent the meaning of the underlying data. Embeddings are frequently used in RAG workflows to find, filter, and represent pertinent data to supplement LLM prompts. Users can search for the most pertinent products using embeddings to power experiences like real-time product recommendations. Finally, application developers typically have experience with and confidence in operational databases to support their enterprise applications.

With the databases you already know and love, especially PostgreSQL, which has become the industry standard for relational databases due to its rich functionality, ecosystem extensions, and vibrant community, we want to make it simple to build enterprise-ready gen AI apps. To start addressing these needs, we introduced support for the well-liked pgvector extension in July in both AlloyDB and Cloud SQL.

In order to meet the demands of a wider range of workloads, AlloyDB AI enhances the fundamental vector support provided by standard PostgreSQL by streamlining the development process and enhancing performance. The end result is a complete method for handling vector embeddings and creating experiences with modern artificial intelligence. With just a few lines of SQL, users can create and query embeddings to find pertinent data; no specialized data stack or data moving is necessary.

What it does

AlloyDB A few new features related to AI are added to AlloyDB to assist developers in integrating their real-time data into next-generation AI applications. These consist of:

• Simple embeddings generation using a PostgreSQL function introduced by AlloyDB AI to create embeddings on your data. You can access Google’s embeddings models with just one line of SQL, including richer local models in Vertex AI and low-latency local models (available as a technology preview in AlloyDB Omni). These models can be used to generate embeddings on-the-fly in response to user inputs and to automatically create embeddings by inferencing in generated columns.
• Thanks to close integrations with the AlloyDB query processing engine, enhanced vector support is possible with up to 10 times faster vector queries than with standard PostgreSQL. Additionally, we present quantization techniques that, when activated, support four times more vector dimensions and a three times space reduction. These techniques are based on Google’s ScaNN technology.
• Integrations with LangChain and Vertex AI Extensions (coming later this year), two components of the AI ecosystem. Finally, we will continue to provide Vertex AI users with the option to call remote models for use-cases like fraud detection that require low latency, high throughput augmented transactions.

By installing the appropriate extensions, these features can be added to any AlloyDB deployment at no additional cost.

Enabling everywhere gen AI apps with AlloyDB Omni

Flexibility and portability were priorities when designing AlloyDB AI. In addition to being PostgreSQL compatible, AlloyDB Omni enables customers to utilize this technology to create enterprise-grade, AI-enabled applications anywhere, including on-premises, at the edge, across clouds, and even on developer laptops. AlloyDB Omni is now in its public preview phase after completing its technology preview.

Clients rely on AlloyDB for their business applications

Customers can keep using AlloyDB as they make use of its AI capabilities because they already have confidence in it for their mission-critical applications. AlloyDB was designed to support high-end applications thanks to its full PostgreSQL compatibility, 99.99% availability, enterprise features like data protection, disaster recovery, and built-in security.

For their most demanding enterprise workloads, the Chicago Mercantile Exchange (CME) Group looks to AlloyDB. A number of databases are currently being converted from Oracle to AlloyDB.

We’re also announcing Duet AI in Database Migration Service if you’re switching from Oracle to AlloyDB. This function offers AI-assisted code conversion to automatically convert Oracle database code, including stored procedures, functions, triggers, packages, and custom PL/SQL code, which could not be converted with conventional translation technologies. Register right away for the preview.

Driving the development of AI

Google Cloud databases offer a platform that enables developers to easily create enterprise-ready generation AI apps using the databases they already know and love. The future of data and AI is bright.

News source: Google cloud