Vector Search Isn’t Enough
summary
Argues that pure vector search misses accuracy and control, and demonstrates a hybrid stack combining Elasticsearch (BM25 + kNN) with Azure OpenAI for RAG-style answers, plus the new Elastic Learned Sparse Encoder for out-of-the-box semantic recall and Reciprocal Rank Fusion to merge rankings. Shows a live demo that indexed conference abstracts, summarizes them via Azure OpenAI (demo cost ~0.224 cents), cites ChatGPT’s ~100M users in days, and plans kNN up to 2,048 dimensions.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 1:32Vector search basics
- 5:21Embedding models
- 7:43Indexing and data types
- 10:26Hybrid search techniques
- 12:02Security and NLP tools
- 14:27Sparse encoder model
- 17:10Reciprocal rank fusion
- 18:56Relevance engine overview
- 20:05Generative AI shift
- 22:03RAG and costs
- 24:26Demo and observability
- 26:48Future AI capabilities
- 28:18Relativity e-discovery
- 30:54AI in legal workflows
- 34:09Case study demo
- 35:58AI ethics and principles
- 38:58Q&A and roadmap
- 41:59Closing remarks
related talks
01h 09mVector Search: Ask Me Anything!
46mAdvanced RAG with Azure AI Search - top questions answered
41mGenAI, vector/semantic/hybrid search, RRF, NLP, (...) and other buzzwords
50mHow to design and build AI applications with vector search using Azure OpenAI & Azure Cosmos DB
47mBoosting AI With OpenSearch: Vector Stores, Similarity Search, and Retrieval-Augme...
01h 16mAzure SQL ❤️ Open AI
37mLexical, Vector & Hybrid Search with Elasticsearch
28mUse OpenSearch as Your Vector Search Database To Optimize GenAI Accuracy!