Google has launched Vertex AI Agent Builder, alongside new open-source language models for the Vertex AI platform. The announcements were made at the Google Cloud Next 2024 event.

Vertex AI is a fully-managed, unified AI development platform for building and using generative AI. It uses Google AI Studio, Agent Builder, and provides 130 foundation models including Gemini 1.5 Pro, all from Vertex AI.

The new Vertex AI Agent Builder, in preview, lets developers build and deploy AI agents using natural language or code. Vertex AI Agent Builder is an expansion and rebranding of Vertex AI Search and Conversation, and the agents can also use open-source frameworks like LangChain on Vertex AI.

The new tool lets you combine prompt-based agent builder tools with pre-built templates. Multiple agents can be stitched together to handle enterprise workflows across multiple channels.

The builder can connect AI agents to users data sources, and can also be used with the Gemini API to make use of results from Google Search and improve the completeness and accuracy of responses. Vertex AI Search's out-of-the-box grounding system is included giving developers the means to ground agents in your enterprise data, and Google says that developers who are looking to build a DIY RAG can can use Google's search components APIs for document processing, ranking, grounded output generation, and performing checks on outputs.

Model grounding is the process of making models that were trained on general 'world data' more specific to the needs of a particular organization. To do this, the responses of foundation models are grounded to reflect the organization's data, to ensure more accurate and complete responses. Google gives the example of grounding foundation model responses in ERP systems, so businesses can create AI agents that provide accurate shipping predictions, or grounding in analyst reports and studies, so compliance can be strengthened by connecting foundation models to contracts.

The Agent Builder comes with connectors so apps can index and make available fresh data from enterprise applications such as JIRA, ServiceNow, and Hadoop.

Vertex AI includes two new text embedding models in public preview: an English-only version and a multilingual version. Google says that compared to previous versions, the new English-language embedding model increased its average score on the MTEB benchmarks, and outperformed all existing entries on MTEB among those with the same embedding size, as well as often outperforming models that are up to 7x larger or have 5x higher dimensions.

More Information

Vertex AI Agent Builder

Related Articles

Google Releases Gemma Open Models

Google Adds Gemini To Bard

Google Rebrands Bard With Subscription

Google Adds Code Generation To Bard

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin.

Comments

Make a Comment or View Existing Comments Using Disqus

or email your comment to: comments@i-programmer.info