Machine Learning
Amazon Bedrock Knowledge Bases
Amazon Bedrock Knowledge Bases manages ingestion, chunking, embeddings, retrieval, and optional response generation over supported data sources and vector stores. It reduces custom RAG plumbing while preserving the need to evaluate source quality and access filters.
Key points
- Data-source synchronization updates the configured vector store
- Retrieval can apply supported metadata filters and reranking options
- Retrieve and RetrieveAndGenerate expose different levels of application control
- Hybrid-search availability depends on the knowledge-base and vector-store configuration
- A guardrail used with a knowledge base evaluates the user input and generated response, not the retrieved references
When to use it
- Ground a support assistant in approved product documents
- Provide citations over an enterprise policy corpus
Exam tips
- Choose Knowledge Bases for managed RAG; build a custom pipeline when chunking or retrieval control exceeds supported options
- A successful sync does not prove recall, freshness, or tenant isolation, so test those separately
- Use retrieve-only evaluation to isolate evidence selection and retrieve-and-generate evaluation to test the complete grounded response