Agentic Hybrid Retrieval: Orchestrating LLM Agents, Hybrid Search & Explainability in OpenSearch
summary
An OpenSearch pipeline uses an LLM agent that generates DSL queries at runtime to choose between hybrid BM25+vector retrieval or both, logs its reasoning for glassbox explainability, and runs on a demo corpus of 100+ books where a vague “books about war” concept query uses vector search while an exact “1994” year query uses keyword search. The prototype wires Llama 3.2 (with GPT‑4o optional), shows higher relevance at the cost of increased P95 latency and tokens, and reduces cost via caching, memory TTLs, and index partitioning.
AI-written summary · about
chapters (11)
- 0:00Agentic hybrid retrieval intro
- 1:33Keyword and vector search
- 5:25Semantic search limits
- 6:29Hybrid search challenge
- 8:38RAG system problems
- 9:43AI agents for dynamic search
- 12:13Agentic search architecture
- 13:34Agents with tools and reasoning
- 15:22Live demo: agentic search
- 18:43Benchmarking agentic search
- 20:53Summary and resources
related talks
51mQuery Understanding With Large Language Models: Techniques and...
30mBeyond Keywords: Building an AI-Powered Search That Reads Your Mind
32mPersonalisation and Specialisation of Search With OpenSearch Agentic Search
49mLLM—Search Engine Killer?
39mExperiments with OpenSearch and AI
38mNatural Language Search-Lexical and Semantic
40mBuilding Intelligent Search Systems With Agentic RAG and OpenSearch
46mLLM—Search Engine Killer?