State of the Art Data Retrieval with Machine Learning & Elasticsearch
summary
Elasticsearch integrates with Microsoft Azure OpenAI to build a retrieval pipeline that creates vector embeddings via model management for 3rd-party transformer models and ML inference pipelines, uses a new pre-trained late interaction model for stronger out‑of‑the‑box relevance, and serves query‑time context windows. Cost estimates show sloppy retrieval can blow up from cents to hundreds of thousands ($225K vs ~$1.9M for 1M queries in examples). A Relativity demo filters 56,000 to 6 documents for targeted review.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:16Relevance engines for generative AI
- 3:48Context window importance
- 7:50Ezra’s document retrieval
- 9:17Semantically similar documents
- 12:10Non-content signals for relevance
- 13:56Domain-specific search context
- 16:48Personalization and security
- 18:57Scaling relevance engines
- 20:28Relativity’s legal tech use case
- 23:21Live demo: AI document discovery
- 26:56Summary
related talks
46mAdvanced RAG with Azure AI Search - top questions answered
50mBuilding AI ready applications
01h 03mDrop the Bass with Embedding and Vectors in Azure AI Search
01h 00m5 Levels of AI in Microsoft Azure
01h 03mDrop the Bass with Embedding and Vectors in Azure AI Search
44mSupercharge your data with Azure AI Search and OpenAI
15mMastering RAG and Vector Search with Elastic
41mResearch in the Age of AI