Vector Search: Ask Me Anything!
summary
Practical Q&A on deploying vector search: benchmarks show a custom APU board can beat Elasticsearch for speed but requires heavy data prep; approximate nearest neighbor methods like HNSW scale well on multi‑core but complicate sharding and exact scoring across shards; combine filtering and ANN in one pass for recall and speed; domain fine‑tuning matters (medical/legal language needs bespoke training or synonyms); long documents prefer chunking or summarization (under ~512 tokens is common); hit highlighting uses fine‑tuned QA models; expect large memory and indexing tradeoffs (colBERT can be hundreds of GB vs BM25 tens of GB).
AI-written summary · about
chapters (20)
- 0:00Intro
- 1:35Dmitry Kan Introduction
- 3:25Max Irwin Introduction
- 6:10Elastic Search Options
- 8:35Is there a Niche
- 10:10Language Models
- 12:40Bird Models
- 14:00Language Detection
- 15:40What do you think
- 17:10Whats next in dense retrieval
- 21:10Where are things heading
- 24:00Prefiltering
- 26:40Content length
- 29:30Content length demo
- 34:35Dmitry on Highlighting
- 36:40Good and Bad Use Cases
- 40:17Dmitrys Thoughts
- 45:13Active Learning
- 47:57Bias
- 50:21Audience Question
related talks
44mVector Search Beyond the Hype: Real-World Implementation and Performance Insights
44mVector Search Isn’t Enough
37mLexical, Vector & Hybrid Search with Elasticsearch
47mLatest innovations from Elasticsearch: Drive speed, scale & relevance (PRT266)
46mNatural Language & Search, Improve your Search w/ Vector capabilities
32mRetrieval Augmentation and Semantic Search at Scale
47mWhy, What, When and How of Running Diff Types of Sear...
59mContextual search with vector search: exploring your options with open source tools