HeatWave GenAI eases development with integrated LLMs, vectors, and more

Organizations building and deploying applications to take advantage of artificial intelligence now have a new and flexible AI-enabled database option. Oracle’s just-released HeatWave GenAI incorporates the tools you need, including large language models (LLMs), an automated vector store, scale-out vector processing, and natural language chat, all in a high performance cloud database platform.

Oracle customers are familiar with HeatWave, which has grown from its MySQL roots into a state-of-the-art multicloud data management system that’s now available on Oracle Cloud Infrastructure (OCI), Amazon Web Services, and Microsoft Azure via Oracle Interconnect for Microsoft Azure. Even before this latest release, Oracle HeatWave supported transactional and analytical workloads and was available with automated machine learning, or AutoML. There’s also Oracle HeatWave Lakehouse, which lets you query data in object stores outside the database.

What’s new and exciting in HeatWave GenAI is that Oracle has integrated core AI capabilities, and it’s done so with performance, automation, and simplicity as top priorities. People familiar with my analysis of the database market know that I put a premium on technologies that minimize complexity and emphasize ease of use. Oracle HeatWave GenAI does just that by incorporating AI functionality that would otherwise require do-it-yourself sourcing and integration of the tech stack. HeatWave processes data in-memory, so it’s fast, too. Oracle benchmarks show HeatWave GenAI to be up to 15 to 30 times faster than Databricks, Google BigQuery, and Snowflake for similarity search at a lower cost.

Valuing simplicity

“One of the big value propositions is that it’s simple,” Nipun Agarwal, Oracle SVP of HeatWave, told me during a demo of HeatWave GenAI. “It doesn’t require AI expertise.”

The net result of having AI-enabling technologies baked into HeatWave GenAI is the ability to skip step-by-step development. Say you want to create a vector store for similarity search. That could require as many as nine distinct actions: discover documents, parse data, extract metadata, segment data, choose embedding model, create vector embeddings, and so on. However, HeatWave GenAI does it all, end-to-end, as an automated process.

“We are replacing these nine steps with a single function,” Agarwal said.

In addition to its built-in LLMs, HeatWave GenAI is integrated with Oracle Cloud Infrastructure (OCI) Generative AI service, giving you access to pretrained, foundational models. Which means that once a vector store is created, HeatWave GenAI’s automation makes it fast and easy to use the vector store with LLMs, again without time-consuming manual steps. Because these components are included, you don’t need AI expertise to build applications. That helps get initiatives up and running quickly.

No need for external LLM services

HeatWave GenAI is best understood through four new capabilities:

• In-database LLMs simplify GenAI development by running popular LLMs, including Mistral and Llama, inside the database. Benefits include ease of implementation, fast performance, and integration with retrieval-augmented generation (RAG). These in-database LLMs are smaller versions of public LLMs, making them an excellent complement to enterprise data for the internal models that many organizations are looking to build.
• HeatWave Vector Store manages vector embeddings without the need for an external, standalone vector database. This marries GenAI capabilities with unstructured corporate data, such as PDFs, Word/PowerPoint documents, and HTML pages. And there may be security and performance advantages because data transformation and processing happen inside the database.
• Scale-out vector processing delivers extremely fast semantic search results. HeatWave enables vector processing to execute at near-memory bandwidth and parallelize across up to 512 HeatWave nodes—hence, “scale out.” It’s the fastest vector processing in the industry, according to Agarwal, and Oracle has the benchmarks to back that up.
• HeatWave Chat adds a graphical interface for developers to ask questions in either natural language or SQL. The database maintains the history of questions posed and prompts used with the LLM, enabling a human-like conversation with follow-up questions. The context of these chats is then maintained, and can be used to help verify the source of answers generated by the LLM.

The beauty of this integrated approach is that developers and IT teams aren’t dependent on external services to build LLM-enabled apps. The necessary components are incorporated into the platform. Thus, it’s a simple, two-step process to build and deploy with LLMs and your business data: One, create a vector store. Two, apply the LLM to your use case.

Enabling new use cases and applications

In addition to its new AI components, HeatWave GenAI takes advantage of existing HeatWave capabilities, including encryption, in-memory query processing, automatic provisioning, and more. HeatWave lets customers run transactions and lakehouse-scale analytics plus automated GenAI and machine learning in one cloud service.

Now, what more can you do with HeatWave GenAI?

Each of the new AI capabilities adds value in its own way. Take HeatWave GenAI’s scale-out vector processing, which employs a mathematically calculated distance function to measure the virtual distance between vector embeddings. In semantic search, data with similar meanings are closer together in the vector space, so a distance function delivers a more precise search result. HeatWave GenAI does so extremely fast.

Here’s another example: In-database LLMs are an attractive alternative to public LLMs offered in an as-a-service model. With HeatWave GenAI, you can run the same LLMs, embeddings, and vector stores across public clouds and OCI regions. This promises greater consistency of GenAI experiences and results. Not to mention that you don’t need to worry about the availability of LLMs in various cloud providers’ data centers.

HeatWave GenAI’s potential is likely even greater when you combine its AI capabilities. For example, the built-in LLMs, vector store, and ML algorithms work together to create more accurate and contextually relevant search results.

Agarwal compares HeatWave AutoML to a “first-pass filter” that skims through documents, then sends a subset of the results to a custom model. In one real-world use case, a food delivery company uses HeatWave AutoML to recommend restaurants based on customers’ previous orders, and HeatWave GenAI to provide additional value by recommending specific dishes to customers based on publicly available menu information housed in the vector store.

This same trio of capabilities—LLMs, vector store, and AutoML—can be used to create business-specific RAG models. An online athletic shoe store could combine public information about trail running with documents and images about the products it sells. Then, if a customer queries, “What shoes are best for rocky trails?” the search results are relevant to both trail running and the store’s inventory.

HeatWave GenAI can also be used for content generation and summarization, turning a product manual into a blog post, for example, and natural language “conversations” with the database via HeatWave Chat.

No AI expertise or data movement required

When does it make sense to choose HeatWave GenAI over another database option?

First and foremost, HeatWave GenAI is an attractive alternative to HeatWave customers, whether they’re using HeatWave MySQL, HeatWave Lakehouse, or HeatWave AutoML. They can easily combine built-in capabilities to help rapidly deliver additional value. Agarwal’s demo highlighted anomaly detection, predictive maintenance, and personalized recommendations.

Organizations looking to build GenAI applications without having to hire AI experts and implement separate vector databases can also benefit from HeatWave GenAI. In addition, because HeatWave GenAI runs on OCI, Azure, and AWS, CIOs and CTOs have a high degree of flexibility as they conceive and implement new AI architectures.

And finally, HeatWave GenAI is an outstanding complement to Oracle’s flagship Oracle Database, including Oracle Autonomous Database and Oracle Database 23ai. In fact, these systems share many of the same advanced AI capabilities, albeit tuned appropriately for each platform.

Oracle is giving its customers choice as they build and deploy a new generation of AI applications. And it’s doing so with the price/performance, innovation, and enterprise capabilities for which Oracle is widely recognized.

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

View more Oracle Connect articles