Analytics
Amazon OpenSearch Service
Amazon OpenSearch Service provides managed search domains and OpenSearch Serverless collections. Both provisioned domains and Serverless vector collections can support semantic, lexical, and hybrid retrieval for RAG.
Key points
- k-NN indexes store and search embedding vectors
- Hybrid search blends vector similarity with keyword relevance
- Network, data-access, and fine-grained permissions are separate control layers
- Mappings, filterable text fields, engines, and backing-store capabilities determine which hybrid and metadata patterns are available
When to use it
- Serve tenant-filtered vector retrieval for a knowledge assistant
- Combine exact product terms with semantic support-search results
Exam tips
- Choose Serverless for collection-level scaling and domains when detailed cluster control is required
- Embedding dimensions and mappings must match; changing models commonly requires a new index and re-embedding
- Measure recall before reranking and relevance after reranking; a larger top-k is not automatically a better or cheaper context