Revolutionizing Search: Unveiling Conversational Search
summary
Shows how to wire OpenSearch to generative models using a configured ML node, a connector to an external LLM endpoint (OpenAI by example), and a RAG pipeline that sends selected context fields plus a conversation memory ID for follow-ups. Demonstrates a working bar-stock demo that returns concrete counts (231 and 578 pints) and a NY population lookup, discusses token limits and cost (~0.2¢ per 1k tokens), and suggests switching to local LLMs and vector retrieval to cut cost and reduce hallucinations.
AI-written summary · about
chapters (16)
- 0:00Conversational search intro
- 1:50Core elements
- 3:35Setting up pipeline
- 4:38Connector config
- 7:25RAG pipeline
- 9:54Indexing data
- 10:59Sample query
- 12:47Memory analysis
- 14:12Cost and privacy
- 15:56Local LLMs
- 16:40Vector embeddings
- 17:24Live demo: bar stock
- 19:07Business potential
- 20:39Q&A: Multi-user convos
- 21:43Q&A: JSON responses
- 23:49Q&A: AI pricing
related talks
21mBuilding RAG With OpenSearch ML Plugins: From PDFs To Voice-Enabled Search
36mBuilding Next-gen Search With OpenSearch
22mAgentic Hybrid Retrieval: Orchestrating LLM Agents, Hybrid Search & Explainability in OpenSearch
37mAgentic AI Meets Search: Real-World Use Cases With...
15mSearch Engineering Reimagined: Leveraging LLM in ML Workflows
28mUse OpenSearch as Your Vector Search Database To Optimize GenAI Accuracy!
21mPersonalize Search Results with OpenSearch Agentic Memory
12mUnconference: Context is all you need?