Main Concept
When building a Knowledge Base in Amazon Bedrock for RAG, you must select a Vector Database to store the embeddings. AWS offers several native options, each suited for different use cases.
AWS Native Options
Amazon OpenSearch Service (Serverless & Managed Cluster)
A search and analytics database designed for real-time similarity queries. It supports storing millions of vector embeddings, scalable index management, and fast nearest-neighbor (kNN) search capability. It is the most common choice for large-scale RAG implementations.
Amazon Aurora PostgreSQL
A relational database, proprietary to AWS, with vector search extensions. It is a good option when your application already uses Aurora and you want to avoid introducing a new service.
Amazon Neptune Analytics
A graph database that enables high-performance graph analytics and graph-based RAG (GraphRAG) solutions. It is useful when relationships between data points matter, such as knowledge graphs or fraud detection scenarios.
Amazon DocumentDB (with MongoDB compatibility)
A document database compatible with MongoDB workloads. Useful when your data is document-oriented and you already work with MongoDB-style queries.
Amazon RDS for PostgreSQL
A managed relational database with vector search support via the pgvector extension. Similar to Aurora PostgreSQL but on standard RDS infrastructure.
Amazon S3 Vectors
Cost-effective and durable vector storage with sub-second query performance. This is a newer option not yet explicitly listed in the current exam guide version, so treat it as supplementary context.
External Options (Beyond Exam Scope)
MongoDB, Redis, and Pinecone are also supported by Amazon Bedrock but are not listed in the exam guide.
📊 Comparison: AWS Vector Databases for RAG
| Service | Primary Strength | Engine / Extension | Best for… | Exam Tip (AI Practitioner) |
|---|---|---|---|---|
| Amazon OpenSearch | Scale & Performance | k-Nearest Neighbor (kNN) | Large-scale, real-time similarity search. | The “default” choice for high-scale RAG. |
| Amazon Aurora | SQL Integration | PostgreSQL + pgvector | Teams already using Aurora who want to avoid new services. | Look for “Relational” + “Vector search”. |
| Amazon Neptune | GraphRAG | Graph Analytics + Vectors | When the relationship between data points is key. | Keywords: “Knowledge Graphs”, “Relationships”. |
| Amazon RDS | Familiarity | PostgreSQL + pgvector | Standard managed PostgreSQL workloads. | Similar to Aurora but for standard RDS. |
| Amazon DocumentDB | JSON Flexibility | MongoDB compatibility | Document-oriented data and MongoDB-style queries. | Keywords: “Document-oriented”, “MongoDB”. |
| Amazon S3 | Cost-Efficiency | Object Storage Vectors | Durable, low-cost vector storage. | Supplementary context (newer option). |
Key Takeaway for the Exam
Focus on identifying these services as valid “Vector Stores” for Amazon Bedrock Knowledge Bases. You don’t need to be a DBA, just know which AWS tools can hold your “embeddings.”
Key Aspects
- The exam expects you to identify which AWS services can serve as vector databases for RAG, not to decide when to use each one.
- The level of detail in this note (kNN, GraphRAG, pgvector) goes beyond what the exam requires, but it is useful context for understanding what differentiates each option.
Summary/Conclusion
AWS offers multiple vector database options for RAG in Amazon Bedrock. For the exam, focus on recognizing the native AWS services: OpenSearch Service, Aurora, Neptune, DocumentDB, and RDS for PostgreSQL. Understanding their general purpose helps contextualize why multiple options exist.