Powering generative AI applications

Defining vectors

Vectors are numerical representations of your data. When vectors represent a form of meaning (semantic), the proximity becomes an indicator for contextual relationship. Your domain-specific data is your most valuable asset for generative AI. By expressing your domain specific data as vectors, you can provide a semantic context for your generative AI applications. Used in this way, such vectors are also called vector embeddings in generative AI.

Use your existing database

Similar to JSON, vectors are a data type that is supported by many OLTP databases. The true value of vectors is realized when you can store, index, retrieve, and search for vector data alongside the non-vector data stored in your databases, allowing for richer understanding of context and enabling more relevant search experiences. If the current database powering your applications supports vectors, you should use that database as it’s already proven to deliver the performance, scalability, availability, and security requirements needed by your application.

AWS has incorporated native support for vectors across a broad set of our databases. For relational use cases, we recommend you use Amazon Aurora PostgreSQL. For non-relational use cases, we recommend you use Amazon MemoryDB, which delivers the fastest vector search performance at the highest recall among popular vector databases on AWS.

image

RAG made easy

Vector search is a critical component of Retrieval-Augmented Generation (RAG), which is used to improve the accuracy of the responses by using your private domain data with large language models (LLMs). It has never been easier to perform RAG. Amazon Bedrock Knowledge Bases is a fully managed RAG capability that automates the data ingestion and runtime orchestration workflows. Amazon Aurora offers 1-click integration with Knowledge Bases so you can easily and securely connect foundation models (FMs) to internal company data sources to deliver more relevant and accurate responses. All it takes is a single click!

GraphRAG

As the expectations for relevancy and explainability of responses have increased, GraphRAG is on the rise. It uses a knowledge graph during the retrieval step in the RAG process. For an automated GraphRAG solution, Amazon Bedrock Knowledge Bases automatically generates graphs (using Amazon Neptune) that links data across multiple sources, including unstructured data. This lets you use your unstructured data to drive value for your business. During the retrieval process, GraphRAG will automatically traverse these graphs to provide more comprehensive, accurate, and explainable responses from LLMs - all with a single API call. With Bedrock Knowledge Bases, the interoperability with Neptune and S3 is built-in resulting in an automated GraphRAG workflow.