Personalize Search Results with OpenSearch Agentic Memory
summary
Claims a working pipeline that personalizes identical queries by having lightweight agents read and write OpenSearch agentic memory (released in 3.3) and enrich queries at request time without retraining models. Uses offline ingestion from UBI clickstream and orders to infer semantic user preferences with an LLM, stores top 5 facts, and serves online with a small model + cached profile, soft filters, neural/hybrid search and a re-ranker. Recommends async memory writes, TTL on memories, and fallback lexical search.
AI-written summary · about
chapters (9)
- 0:00Introduction to search personalization
- 1:03Why traditional personalization fails
- 3:55Introducing agentic query personalization
- 5:40OpenSearch agentic memory types
- 7:25Storing user preferences in memory
- 9:31Offline ingestion pipeline
- 12:43Online search pipeline
- 15:35Production recommendations
- 18:07Live demo walkthrough
related talks
32mOpenSearch Agentic Memory: Intelligent Context for AI Agents
32mPersonalisation and Specialisation of Search With OpenSearch Agentic Search
40mAgentic Relevance Tuning With OpenSearch: Faster, Easier, and Higher Quality Search at Scale
22mAgentic Hybrid Retrieval: Orchestrating LLM Agents, Hybrid Search & Explainability in OpenSearch
41mIntegrating Realtime User Behavioral Embeddings Into OpenSearch
40mBuilding Intelligent Search Systems With Agentic RAG and OpenSearch
26mRevolutionizing Search: Unveiling Conversational Search
38mAI-Ready Data Foundation Using Amazon OpenSearch