Amazon OpenSearch Service
Search, log analytics, and vector database for RAG.
❓ What is it?
A managed OpenSearch cluster (and serverless option) for full-text search and log analytics that now doubles as a vector database: k-NN indexes store embeddings for semantic search at scale.
💡 Why does it exist?
RAG needs somewhere to store and search embeddings fast. OpenSearch combines classic keyword search with vector similarity (hybrid search), which beats either alone for retrieval quality.
⏱️ When should you use it?
Use it as the vector store for Bedrock Knowledge Bases (its serverless flavour is the default), for semantic product search, and for log/observability analytics.
🗺️ Where does it fit?
In the retrieval layer of GenAI architectures: ingestion pipelines write embeddings; queries embed the question and pull nearest neighbours; Bedrock composes the grounded answer.
🔌 How do you integrate it?
Create a domain or serverless collection, define a k-NN index with your embedding dimensions, ingest vectors alongside text/metadata, and issue hybrid (keyword + vector) queries.
🧩 Commonly integrated with
🎯 Exam angle (AIF-C01)
- Vector database on AWS in exam answers: OpenSearch (incl. serverless) — the default Knowledge Bases store; pgvector on Aurora/RDS, Neptune, DocumentDB, MemoryDB also support vectors.
- Hybrid search (keyword + semantic) is its differentiator for retrieval-quality questions.