Unlocking Next-gen Semantic Search With OpenSearch 3.x
summary
OpenSearch 3.x demonstrates concrete recipes for billion‑scale vectors: GPU indexing via Cagra, Faiss and JVector engines, and HNSW/IVF ANN with Radial Search and Maximal Marginal Relevance (MMR) for diversity. Practical optimizations include derived source and disk‑based search to avoid keeping 4 TB RAM for 1B vectors at 1,024 dims; product quantization can shrink that to ~370 GB and binary quantization to ~268 GB. Also: Memory Optimized Search, asymmetric distance computation, random rotation, and a Workflow Builder for chaining ingest and rerank pipelines.
AI-written summary · about
chapters (26)
- 0:00OpenSearch semantic search intro
- 1:06What is semantic search
- 3:15Semantic search workflow
- 5:49Embedding models in OpenSearch
- 6:55OpenSearch vector field types
- 7:57KNN vector field deep dive
- 12:14ANN algorithms in OpenSearch
- 13:37Optimizing IVF and HNSW
- 17:35Sparse vectors applications
- 19:19Semantic field type config
- 20:43Chunking large documents
- 23:50OpenSearch search modes
- 26:17Radial and hybrid search
- 27:20Rescoring for recall
- 28:24Maximal marginal relevance
- 30:08Billion-scale vector search
- 31:11GPU acceleration indexing
- 33:58Storage optimizations vectors
- 36:51Quantization memory reduction
- 40:48Advanced quantization 3.x
- 41:30ML model memory allocation
- 42:31Disk-based hybrid search
- 43:34Precision and recall techniques
- 44:58OpenSearch workflow builder
- 45:41Semantic search future
- 46:03Q&A parameter tuning
related talks
55mVector Search Made Simple: Getting Started with OpenSearch for AI Applications
43mOpenSearch and the Path to a Trillion-scale Vector Database
26mOpenSearch and the Path to a Trillion-s...
32mRetrieval Augmentation and Semantic Search at Scale
22mVector Search Made Simple: Getting Started With OpenSearch for AI Applications
30mVector Search Made Simple: Getting Started with OpenSearch for AI Applications
46mNatural Language & Search, Improve your Search w/ Vector capabilities
36mWhen Semantic Search Breaks in your RAG systems: RAM Walls, Silent Failures, and the Architecture Decisions