Search: A New Era
summary
Elasticsearch now stores and searches dense embeddings, supports 4,000 dimensions (tradeoff: memory and slower indexing), and uses HNSW + brute force switching for fast nearest‑neighbor queries; commercial inference pipelines (commercial) can generate text embeddings in‑cluster. The demo indexes audio embeddings and matches a hummed tune — the music humming demo returns sensible nearest neighbors. Hybrid scoring combines BM25, sparse text expansion and vector KNN; rank fusion (RRF with constant 60) normalizes ranks. Practical tips: exclude vectors from _source, prefer bytes or fewer dims, avoid searches during heavy indexing.
AI-written summary · about
chapters (10)
- 0:00Modern search evolution
- 3:59Traditional term-based limits
- 7:53Vector embeddings and semantic search
- 12:27Vector search in Elasticsearch
- 18:42Vector search performance tips
- 23:38Term and vector search combined
- 24:23Music search with vectors
- 30:43Elasticsearch with LLMs
- 34:06Semantic search with Elasticsearch
- 37:20Future of search
related talks
42mResearch in the Age of AI
47mLatest innovations from Elasticsearch: Drive speed, scale & relevance (PRT266)
45mBehind the hype: managing billion-scale embeddings in Elasticsearch and OpenSearch
37mLexical, Vector & Hybrid Search with Elasticsearch
46mNatural Language & Search, Improve your Search w/ Vector capabilities
01h 09mVector Search: Ask Me Anything!
13mA Beginner’s Guide to Semantic Search
16mDense Vectors in Elasticsearch