Why, What, When and How of Running Diff Types of Sear...
summary
Compares running vector, semantic, neural, conversational and multimodal search on OpenSearch between managed and self‑hosted setups with concrete numbers and tradeoffs: 50 TB vs 100 TB sizing, an estimated initial ~$1M procurement for on‑prem hardware, and per‑instance pay‑as‑you‑go for managed. Recommends plugins and patterns: approximate KNN, separate indexes for vectors/lexical, reranking or script_score for hybrid results, GPUs/high IO for multimodal, monitoring for index bloat and token explosion, and when autoscaling or serverless fits operational versus BI workloads.
AI-written summary · about
chapters (21)
- 0:00Introduction
- 1:03Search relevance importance
- 2:52Choosing search types
- 7:49Semantic search example
- 9:13Business intelligence with semantic
- 9:37Conversational search SLA
- 10:42Hybrid search log analysis
- 11:24Multimodal search troubleshooting
- 12:46OpenSearch managed vs self-hosted
- 13:32Infrastructure needs per search type
- 16:47Cost and scalability comparison
- 20:23Semantic search best practices
- 22:31Hybrid search ranking techniques
- 25:20Multimodal search indexing
- 27:49Neural search tuning
- 29:57Conversational search context
- 32:01Key takeaways
- 34:08Managed vs self-hosting
- 35:15Enriching data with annotations
- 37:03Global intelligence sharing
- 38:09Q&A
related talks
48mUnlocking Next-gen Semantic Search With OpenSearch 3.x
01h 09mVector Search: Ask Me Anything!
32mRetrieval Augmentation and Semantic Search at Scale
46mNatural Language & Search, Improve your Search w/ Vector capabilities
26mOpenSearch and the Path to a Trillion-s...
30mSearch on Steroids — A Story of Semantics
41mText to Context: How We Introduced Hybrid Search
54mThe Debate Returns (with more vectors): Which Search Engine?