Google has introduced a new feature in the realm of LLM, called Vertex AI Agent Builder, which allows for the development of services in LLM without having to write direct code (except for external integrations).
This service combines several capabilities, including:
– Data store creation for documents to be used, under the service named Vertex AI Search
– Function calling with Open API specs and descriptions
– Answer grounding system
– Final answer testing system for Agents
– Data connectors to link external services such as ServiceNow and Salesforce
The Vertex AI Search tool is a ready-made Retrieval Augmented Generation (RAG) model that comes equipped with multiple capabilities, but Google has now released internal features separately, including:
– Grounding API: to verify if answers match the data sources, even in cases of multiple sources
– Ranking API: to rank data that corresponds to the question, more accurate than using vector search alone but suited for smaller data sets
– Document AI Layout Parser: a service that transforms documents into a format readable by LLM, removing headers/footers and rearranging pages
– Grounded Generation API: to generate answers from questions and data sources
For those looking to customize further, LangChain on Vertex AI offers an even higher level of refinement.
Source: Google Cloud Blog
TLDR: Google introduces Vertex AI Agent Builder for LLM development, providing various services without the need for direct coding, including data storage, function calling, answer verification, and external service connectivity. Additional tools such as Grounding API, Ranking API, Document AI Layout Parser, and Grounded Generation API enhance the capabilities for tailored solutions.
Leave a Comment