Oracle announces industry first in-database LLMs and an automated in-database vector store with HeatWave GenAI

Oracle have announced the general availability of HeatWave GenAI, which includes the industry’s first in-database large language models (LLMs), an automated in-database vector store, scale-out vector processing, and the ability to have contextual conversations in natural language informed by unstructured content. These new capabilities enable customers to bring the power of generative AI to their enterprise data – without requiring Artificial Intelligence (AI) expertise or having to move data to a separate vector database. HeatWave GenAI is available immediately in all Oracle Cloud regions, Oracle Cloud Infrastructure (OCI) Dedicated Region, and across clouds at no extra cost to HeatWave customers.

With HeatWave GenAI, developers can create a vector store for enterprise unstructured content with a single SQL command, using built-in embedding models. Users can perform natural language searches in a single step using either in-database or external LLMs. Data doesn’t leave the database and, due to HeatWave’s extreme scale and performance, there is no need to provision GPUs. As a result, developers can reduce application complexity, increase performance, improve data security, and lower costs.

“HeatWave’s stunning pace of innovation continues with the addition of HeatWave GenAI to existing built-in HeatWave capabilities: HeatWave Lakehouse, HeatWave Autopilot, HeatWave AutoML, and HeatWave MySQL,” said Edward Screven, chief corporate architect, Oracle.

“Today’s integrated and automated AI enhancements allow developers to build rich generative AI applications faster, without requiring AI expertise or moving data. Users now have an intuitive way to interact with their enterprise data and rapidly get the accurate answers they need for their businesses.”

“HeatWave GenAI makes it extremely easy to take advantage of generative AI,” said Vijay Sundhar, Chief Executive Officer, SmarterD. “The support for in-database LLMs and in-database vector creation leads to a significant reduction in application complexity, predictable inference latency, and most of all, no additional cost to us to use the LLMs or create the embeddings. This is truly the democratisation of generative AI and we believe it will result in building richer applications with HeatWave GenAI and significant gains in productivity for our customers.”

New automated and built-in generative AI features include: 

• In-database LLMs simplify the development of generative AI applications at a lower cost. Customers can benefit from generative AI without the complexity of external LLM selection and integration, and without worrying about the availability of LLMs in various cloud providers’ data centres. The in-database LLMs enable customers to search data, generate or summarise content, and perform retrieval-augmented generation (RAG) with HeatWave Vector Store. In addition, they can combine generative AI with other built-in HeatWave capabilities such as AutoML to build richer applications. HeatWave GenAI is also integrated with the OCI Generative AI service to access pre-trained, foundational models from leading LLM providers.
• Automated in-database Vector Store enables customers to use generative AI with their business documents without moving data to a separate vector database and without AI expertise. All the steps to create a vector store and vector embeddings are automated and executed inside the database, including discovering the documents in object storage, parsing them, generating embeddings in a highly parallel and optimised way, and inserting them into the vector store making HeatWave Vector Store efficient and easy to use. Using a vector store for RAG helps solve the hallucination challenge of LLMs as the models can search proprietary data with appropriate context to provide more accurate and relevant answers.
• Scale-out vector processing delivers very fast semantic search results without any loss of accuracy. HeatWave supports a new, native VECTOR data type and an optimised implementation of the distance function, enabling customers to perform semantic queries with standard SQL. In-memory hybrid columnar representation and the scale-out architecture of HeatWave enable vector processing to execute at near-memory bandwidth and parallelize across up to 512 HeatWave nodes. As a result, customers get their questions answered rapidly. Users can also combine semantic search with other SQL operators to, for example, join several tables with different documents and perform similarity searches across all documents.
• HeatWave Chat is a Visual Code plug-in for MySQL Shell which provides a graphical interface for HeatWave GenAI and enables developers to ask questions in natural language or SQL. The integrated Lakehouse Navigator enables users to select files from object storage and create a vector store. Users can search across the entire database or restrict the search to a folder. HeatWave maintains context with the history of questions asked, citations of the source documents, and the prompt to the LLM. This facilitates a contextual conversation and allows users to verify the source of answers generated by the LLM. This context is maintained in HeatWave and is available to any application using HeatWave.

 Vector Store Creation and Vector Processing Benchmarks

Creating a vector store for documents in PDF, PPT, WORD, and HTML formats is up to 23X faster with HeatWave GenAI and 1/4th the cost of using Knowledge base for Amazon Bedrock.

As demonstrated by a third-party benchmark using a variety of similarity search queries on tables ranging from 1.6GB to 300GB in size, HeatWave GenAI is 30X faster than Snowflake and costs 25 percent less, 15X faster than Databricks and costs 85 percent less, and 18X faster than Google BigQuery and costs 60 percent less.

A separate benchmark reveals that vector indexes in Amazon Aurora PostgreSQL with pgvector can have a high degree of inaccuracy and can yield incorrect results. In contrast, HeatWave similarity search processing always provides accurate results, has predictable response time, is performed at near memory speed, and is up to 10X-80X faster than Aurora using the same number of cores.

“We are thrilled to continue our strong collaboration with Oracle to deliver the power and productivity of AI with HeatWave GenAI for critical enterprise workloads and data sets,” said Dan McNamara, senior vice president and general manager, Server Business Unit, AMD. “The joint engineering work undertaken by AMD and Oracle is enabling developers to design innovative enterprise AI solutions by leveraging HeatWave GenAI powered by the core density and outstanding price-performance of AMD EPYC processors.”

Additional Customer and Analyst Commentary on HeatWave GenAI

“We heavily use the in-database HeatWave AutoML for making various recommendations to our customers,” said Safarath Shafi, chief executive officer, EatEasy. “HeatWave’s support for in-database LLMs and in-database vector store is differentiated and the ability to integrate generative AI with AutoML provides further differentiation for HeatWave in the industry, enabling us to offer new kinds of capabilities to our customers. The synergy with AutoML also improves the performance and quality of the LLM results.”

“HeatWave in-database LLMs, in-database vector store, scale-out in-memory vector processing, and HeatWave Chat, are very differentiated capabilities from Oracle that democratise generative AI and make it very simple, secure, and inexpensive to use,” said Eric Aguilar, founder, Aiwifi. “Using HeatWave and AutoML for our enterprise needs has already transformed our business in several ways, and the introduction of this innovation from Oracle will likely spur growth of a new class of applications where customers are looking for ways to leverage generative AI on their enterprise content.”

“HeatWave’s engineering innovation continues to deliver on the vision of a universal cloud database,” said Holger Mueller, vice president and principal analyst, Constellation Research. “The latest is generative AI done ‘HeatWave style’—which includes the integration of an automated, in-database vector store and in-database LLMs directly into the HeatWave core. This enables developers to create new classes of applications as they combine HeatWave elements. For example, they can combine HeatWave AutoML and HeatWave GenAI in a fraud detection application that not only detects suspicious transactions—but also provides an understandable explanation. This all runs in the database, so there’s no need to move data to external vector databases, keeping the data more secure. It also makes HeatWave GenAI highly performant at a fraction of the cost as demonstrated in competitive benchmarks.”