Unconference: Context is all you need?
summary
Claim: context selection — not bigger models — fixes hallucination and cost for production GenAI. Demonstration-level pattern: ingest core documents into OpenSearch as a primary vectorstore + online lookups, run retrieval first, then fetch supporting records from S3/DBs at query time, assemble context, and call the LLM. Experiments show large token windows blow cost and induce hallucination; OpenSearch handles semantic retrieval while a separate graph store captures multi-file relationships. This is a practical RAG pipeline design that reduces generator load and latency.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:06GenAI context challenges
- 2:54Knowledge graphs and semantic search
- 4:04Context propagation questions
- 6:14Retrieval strategies for RAG
- 9:30Graph data for context building
- 10:35OpenSearch and graph data roles
- 10:55Community effort for GenAI context
related talks
26mRevolutionizing Search: Unveiling Conversational Search
49mContext is King
01h 03mAgentic Search for Context Engineering
20mThe Future of Enterprise AI Is Context Infrastructure, Not Better Models
22mAgentic Hybrid Retrieval: Orchestrating LLM Agents, Hybrid Search & Explainability in OpenSearch
15mUnderstanding Context Engineering for Agentic Systems
31mBuild Intelligent AI Agents With Context Engineering
32mOpenSearch Agentic Memory: Intelligent Context for AI Agents