Query Understanding With Large Language Models: Techniques and...
summary
Shows an OpenSearch pipeline that uses LLMs (via Strands agents and Amazon Bedrock in the demo) to do query expansion and filter extraction, turning “black shoes” into enriched text + explicit filters and raising cosine scores from ~0.79 to ~0.82 in the demo. Architecture uses a default fast path with a 200–300 ms e‑commerce SLA and an LLM-first path with semantic fallback plus semantic caching to cut inference cost. Reindexing required when changing embeddings; hybrid search and observability are recommended.
AI-written summary · about
chapters (9)
- 0:00Query understanding with LLMs
- 1:46Search technologies limits
- 6:59Query rewriting techniques
- 9:04User domain context
- 15:50Search personas
- 22:55LLM query expansion
- 27:58Demo setup
- 37:49Production deployment
- 41:42Q&A on filters
related talks
30mBeyond Keywords: Building an AI-Powered Search That Reads Your Mind
22mAgentic Hybrid Retrieval: Orchestrating LLM Agents, Hybrid Search & Explainability in OpenSearch
38mNatural Language Search-Lexical and Semantic
24mConversational Search with LLMs
44mRediscover your keyword search: Expand, Enrich and Rewrite
49mLLM—Search Engine Killer?
39mExperiments with OpenSearch and AI
01h 30mUsing LLMs to improve your Search Engine