Building advanced RAGs with Elasticsearch OpenSearch
summary
Walks through building a production RAG using Elasticsearch/OpenSearch, showing concrete steps: chunking and embeddings for clean vectors, indexing tens to hundreds of thousands of docs, then running parallel hybrid search plus RF (semantic + keyword) with metadata filters, applying a lightweight re‑ranking model and time decay to boost fresh content. Presents concrete failure modes (wrong regional policy returned), fixes that reduced hallucination, and recommends human + retrieval evaluation and metricized golden sets.
AI-written summary · about
chapters (12)
- 0:00Introduction and why RAG matters today
- 2:30Evolution of LLM applications: chatbots, workflows, RAG, agents
- 6:00What RAG actually is (and why it’s not “dead”)
- 9:00Anatomy of a RAG system (embeddings, vector DB, retrieval)
- 13:00Choosing vector databases (Elasticsearch, OpenSearch, others)
- 16:00Common RAG failures and bad retrieval examples
- 19:00Fixing preprocessing: chunking, cleaning, embeddings
- 22:00Keyword vs semantic search trade-offs
- 25:00Hybrid search and metadata filtering strategies
- 28:00Re-ranking models and improving relevance
- 30:00Time decay and advanced ranking techniques
- 31:30Evaluation strategies and production best practices
related talks
45mMaking semantic search & RAG real: How to build a prod-ready app
20mChallenges in Building RAG Systems (and Solutions Using OpenSearch)
21mSmart Search with RAG: Elasticsearch Meets Language Models
46mBlow past benchmarks using vector search and RAG with Elasticsearch
34mImplementing an open-source RAG with OpenSearch
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
19mImproving RAG Systems - Key Challenges and Solutions
02h 47mFrom naive to advanced RAG: the complete guide