Boosting AI With OpenSearch: Vector Stores, Similarity Search, and Retrieval-Augme...
summary
OpenSearch is shown handling dense and sparse embeddings with ML Commons model serving, the neural plugin injection/search pipelines and a serverless vector engine; demos index product catalogs and multimodal queries (image+text) and return more relevant results than lexical search. Guidance includes choosing dimensions (300 or 768), engine (HNSW or IVF), shard sizing (~30–40 GB), and quantization for scaling from millions to billions of vectors. Amazon Personalize and LTR rerankers plug into retrieval/RAG flows and the demos validate the end-to-end pipelines.
AI-written summary · about
chapters (16)
- 0:00Vector search with OpenSearch
- 1:06Vector search use cases
- 6:07Lexical vs semantic search
- 8:55Personalized ranking
- 12:30Bringing models to OpenSearch
- 16:19Semantic search in OpenSearch
- 22:21Vector storage algorithms
- 25:59Hybrid search combining methods
- 29:15Multimodal search
- 31:22Sparse vector search
- 34:14Vector search architecture
- 38:07Vector quantization techniques
- 40:37Indexing and search performance
- 43:01Benchmarking vector search
- 43:47Live demo: vector search
- 46:44Q&A
related talks
28mUse OpenSearch as Your Vector Search Database To Optimize GenAI Accuracy!
38mA Visual AI Search Flow Builder for OpenSearch: Build GenAI and Vector...
44mRediscover your keyword search: Expand, Enrich and Rewrite
26mOpenSearch: Conversations with Context
38mAI-Ready Data Foundation Using Amazon OpenSearch
38mNatural Language Search-Lexical and Semantic
17mVectors Vs. Hallucinations: OpenSearch's GenAI Survival Kit
25mRedefining Search: Vector Search with OpenAI and MongoDB