Making semantic search & RAG real: How to build a prod-ready app
summary
Elastic shows a production path combining Elasticsearch + Amazon Bedrock models and Lucene upgrades to serve RAG: vector search uses HNSW for ANN, passage vectors (now in 8.11) chunk long docs, SIMD CPU paths and int8 quantization cut memory by ~4x, and cross-cluster replication adds DR while RBAC secures results. Reciprocal rank fusion blends BM25 and vector scores for relevance, and native embedding hooks (Hugging Face, LangChain) let teams fine-tune domain models and run inference without stitching microservices.
AI-written summary · about
chapters (5)
- 0:00Introduction to generative AI
- 2:14Foundation models and RAG
- 6:34Elasticsearch vector features
- 15:44Advancements in vector search
- 34:45Production use case in commerce
related talks
37mBuilding advanced RAGs with Elasticsearch OpenSearch
21mSmart Search with RAG: Elasticsearch Meets Language Models
46mBlow past benchmarks using vector search and RAG with Elasticsearch
59mThe Evolution of RAG
02h 47mFrom naive to advanced RAG: the complete guide
20mChallenges in Building RAG Systems (and Solutions Using OpenSearch)
19mRAG in 2025: State of the Art and the Road Forward
34mImplementing an open-source RAG with OpenSearch