At the Google Cloud Next conference today, Google will announce a series of AI-powered updates across its portfolio, including its database platforms.
Among the AI focused database announcements is the introduction of AlloyDB AI, which brings vector embeddings to the PostgreSQL compatible cloud database. The new vector embeddings will also be part of the AlloyDB Omni service which is entering public preview today, enabling users to run AlloyDB outside of the Google Cloud.
AlloyDB was first announced as a preview by Google in May 2022, providing both transactional and analytics capabilities with a PostgreSQL based database. The AlloyDB Omni platform was initially detailed by Google in March 2023, opening up the database to wider deployment options.
AI will also help to enable database migrations from the Oracle database to AlloyDB, with the new Duet AI capability in the Google Database Migration Service. Beyond AlloyDB, Google is also introducing the new Cloud Spanner Data Boost capability that will enable data sorted in the Cloud Spanner database to be more easily queried with Google BigQuery. Duet AI is also making its way into Cloud Spanner to help enable natural language queries for data operations.
“We really see databases as helping to bridge the gap between large language models (LLMs) and AI apps,” Andi Gutmans, VP and GM for databases at Google, told VentureBeat. “Customers, especially enterprise customers, really like the ChatGPT experience, but ultimately they can’t have something that is too creative, and they really need to anchor their generative AI apps in the actual enterprise data.”
Vector enabled databases are increasingly critical to enabling databases to be data stores for AI applications.
While there are purpose built vector databases like Pinecone and milvus, existing database platforms such as PostgreSQL have also increasingly made efforts to support vectors. In PostgreSQL, the open-source pgvector technology is often used in the open-source database to support vectors. Some vendors such as Neon, which is a PostgreSQL compatible cloud database, have gone beyond pgvector, with Neon developing its own pg_embedding approach to supporting vectors in PostgreSQL.
Gutmans explained that with AlloyDB, Google is providing AI is a ‘superset’ of capabilities on top of pgvector. For one, the vector capabilities have been integrated deeply into the AlloyDB query processing engine.
“We’re probably smarter in how we execute the queries and how we optimize the queries,” said Gutmans.
The other key element is added vector quantization support. Getmans explained that quantization enables AlloyDB users to significantly reduce vectors’ resource footprint in a running database, which helps improve and reduce storage costs.
Beyond just boosting pgvector, Gutmans emphasized that Google’s goal is to make it easier for developers to bring LLMs and enterprise data together.
AlloyDB AI integrates an easy way for developers to generate vector embeddings in several approaches. One approach is via an integration with Google’s Vertex AI to create vector embeddings. Additionally, Gutmans noted that Google is integrating a series of very lightweight embeddings models into the database. Integration with the open source LangChain technology is also part of the rollout, with the goal to help developers pull together data for AI powered applications.
“You should really think about [AlloyDB} as being all the different capabilities that developers need to be successful and bridging the gap between the data and LLMs,” said Gutmans.
PostgreSQL — and by extension, databases such as AlloyDB based on it — have long been positioned as potential alternatives to the Oracle database.
Google has been iterating on its own database migration service for its databases over the last several years. The database migration service aims to automatically map an existing Oracle database and its functions into an AlloyDB deployment. Gutmans explained that the existing technology is a rules-based model that meets many requirements, but it doesn’t solve for all use cases. That’s where the new Duet AI in the database migration service fits in.
The Duet AI in the database migration service enables developers to provide a prompt with manual hints on how they want to migrate certain parts of their Oracle database stored procedures. Gutmans said that Duet AI uses an LLM to generate the necessary code that can run across a cluster.
“There’s only so much you can do with a rules-based engine to migrate Oracle stored procedures to PostgreSQL,” said Gutmans. “Duet AI is basically an AI system for folks doing code conversion for that last mile that we couldn’t actually convert automatically.”
Head over to our on-demand library to view sessions from VB Transform 2023. Register Here
At the Google Cloud Next conference today, Google will announce a series of AI-powered updates across its portfolio, including its database platforms.
Among the AI focused database announcements is the introduction of AlloyDB AI, which brings vector embeddings to the PostgreSQL compatible cloud database. The new vector embeddings will also be part of the AlloyDB Omni service which is entering public preview today, enabling users to run AlloyDB outside of the Google Cloud.
AlloyDB was first announced as a preview by Google in May 2022, providing both transactional and analytics capabilities with a PostgreSQL based database. The AlloyDB Omni platform was initially detailed by Google in March 2023, opening up the database to wider deployment options.
Easier queries with natural language
AI will also help to enable database migrations from the Oracle database to AlloyDB, with the new Duet AI capability in the Google Database Migration Service. Beyond AlloyDB, Google is also introducing the new Cloud Spanner Data Boost capability that will enable data sorted in the Cloud Spanner database to be more easily queried with Google BigQuery. Duet AI is also making its way into Cloud Spanner to help enable natural language queries for data operations.
Event
VB Transform 2023 On-Demand
Did you miss a session from VB Transform 2023? Register to access the on-demand library for all of our featured sessions.
“We really see databases as helping to bridge the gap between large language models (LLMs) and AI apps,” Andi Gutmans, VP and GM for databases at Google, told VentureBeat. “Customers, especially enterprise customers, really like the ChatGPT experience, but ultimately they can’t have something that is too creative, and they really need to anchor their generative AI apps in the actual enterprise data.”
Vectors in AlloyDB is more than just pgvector
Vector enabled databases are increasingly critical to enabling databases to be data stores for AI applications.
While there are purpose built vector databases like Pinecone and milvus, existing database platforms such as PostgreSQL have also increasingly made efforts to support vectors. In PostgreSQL, the open-source pgvector technology is often used in the open-source database to support vectors. Some vendors such as Neon, which is a PostgreSQL compatible cloud database, have gone beyond pgvector, with Neon developing its own pg_embedding approach to supporting vectors in PostgreSQL.
Gutmans explained that with AlloyDB, Google is providing AI is a ‘superset’ of capabilities on top of pgvector. For one, the vector capabilities have been integrated deeply into the AlloyDB query processing engine.
“We’re probably smarter in how we execute the queries and how we optimize the queries,” said Gutmans.
The other key element is added vector quantization support. Getmans explained that quantization enables AlloyDB users to significantly reduce vectors’ resource footprint in a running database, which helps improve and reduce storage costs.
Alloy DB AI helps developers create vector embeddings
Beyond just boosting pgvector, Gutmans emphasized that Google’s goal is to make it easier for developers to bring LLMs and enterprise data together.
AlloyDB AI integrates an easy way for developers to generate vector embeddings in several approaches. One approach is via an integration with Google’s Vertex AI to create vector embeddings. Additionally, Gutmans noted that Google is integrating a series of very lightweight embeddings models into the database. Integration with the open source LangChain technology is also part of the rollout, with the goal to help developers pull together data for AI powered applications.
“You should really think about [AlloyDB} as being all the different capabilities that developers need to be successful and bridging the gap between the data and LLMs,” said Gutmans.
AI power comes to database migration
PostgreSQL — and by extension, databases such as AlloyDB based on it — have long been positioned as potential alternatives to the Oracle database.
Google has been iterating on its own database migration service for its databases over the last several years. The database migration service aims to automatically map an existing Oracle database and its functions into an AlloyDB deployment. Gutmans explained that the existing technology is a rules-based model that meets many requirements, but it doesn’t solve for all use cases. That’s where the new Duet AI in the database migration service fits in.
The Duet AI in the database migration service enables developers to provide a prompt with manual hints on how they want to migrate certain parts of their Oracle database stored procedures. Gutmans said that Duet AI uses an LLM to generate the necessary code that can run across a cluster.
“There’s only so much you can do with a rules-based engine to migrate Oracle stored procedures to PostgreSQL,” said Gutmans. “Duet AI is basically an AI system for folks doing code conversion for that last mile that we couldn’t actually convert automatically.”
VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.
Author: Sean Michael Kerner
Source: Venturebeat
Reviewed By: Editorial Team
