Amazon Bedrock Knowledge Bases has broadened its hybrid search capabilities by integrating support for Amazon Aurora PostgreSQL and MongoDB Atlas vector stores. This enhancement allows developers to leverage hybrid search—a combination of semantic and keyword-based search methods—to improve the relevance of results in Retrieval Augmented Generation (RAG) applications.
Understanding Hybrid Search in RAG Applications
Hybrid search synergizes semantic search, which interprets the contextual meaning of queries, with traditional keyword search that identifies exact term matches. This dual approach ensures more comprehensive and accurate retrieval of information, particularly beneficial for queries requiring both conceptual understanding and specific keyword identification. For instance, in customer service chatbots, hybrid search can effectively handle inquiries that involve both general topics and specific product names or terms.
Previous Limitations and New Integrations
Initially, hybrid search within Amazon Bedrock Knowledge Bases was exclusive to Amazon OpenSearch Serverless. The recent integration of Amazon Aurora PostgreSQL and MongoDB Atlas as supported vector stores marks a significant expansion, providing developers with more flexibility and options for their RAG implementations.
- Amazon Aurora PostgreSQL: Now available as a quick-create vector store, Aurora PostgreSQL simplifies the setup process for developers. By utilizing the pgvector extension, it enables efficient storage and querying of vector embeddings, facilitating seamless integration with Bedrock Knowledge Bases.
- MongoDB Atlas: The integration with MongoDB Atlas allows developers to utilize its vector search capabilities within Bedrock Knowledge Bases. This enables the development of generative AI applications that can process and retrieve information from operational data stored in MongoDB, enhancing the accuracy and relevance of AI-generated responses.
Availability and Getting Started
Hybrid search with Aurora PostgreSQL is accessible in all AWS Regions where Bedrock Knowledge Bases is available, excluding Europe (Zurich) and GovCloud (US) Regions. For MongoDB Atlas, hybrid search support is currently available in the US East (N. Virginia) and US West (Oregon) AWS Regions.
Developers can enable hybrid search through the Knowledge Base APIs or via the Amazon Bedrock console by selecting hybrid search as the preferred search option. This advancement empowers developers to build more robust and contextually aware generative AI applications by leveraging the combined strengths of semantic and keyword search methodologies.
For more detailed information and to get started, refer to the Amazon Bedrock Knowledge Bases documentation and the respective integration guides for Amazon Aurora PostgreSQL and MongoDB Atlas.
