Advancing Multi-Modal Search Capabilities in Search Pipeline
summary
Demonstrates ML inference processors inserted at ingest/search/response stages of an OpenSearch pipeline that generate embeddings, rewrite queries, and rerank responses; demo ingests ~2,000 product records, converts images to base64, calls a remote model (Amazon Bedrock Titan embeddings) to produce 1,024‑dimension vectors and stores them in a KNN HNSW 1,024-dimension index; shows text, image and cross‑lingual queries returning relevant results, plus conversational RAG via agents and plans for a model‑connector (MCP) server and async/streaming inference.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 0:28What is Open Search
- 1:09Basic Capabilities of Open Search
- 1:51Open Search ML
- 2:21ML Commons
- 4:40Agenda
- 7:00ML Inference Processor
- 7:48ML Inference Processor Diagram
- 10:14ML Inference Processor Integration
- 11:10Demo
- 16:59Results
- 21:50Questions
related talks
46mNatural Language & Search, Improve your Search w/ Vector capabilities
41mText to Context: How We Introduced Hybrid Search
22mBuilding a Multimodal Search Engine
48mUnlocking Next-gen Semantic Search With OpenSearch 3.x
27mMachine Learning for Advanced AI Search
01h 30mBuilding a Search Engine
36mSearch Complete
01h 08mGoing multi-modal