Orchestrating Multimodal RAG With Agentic Workflows With OpenSearch
summary
Claims a working Multimodal Agentic RAG that combines OpenSearch (including OpenSearch Serverless), DeepSeek-R1 reasoning, and custom-tuned embeddings to ground agents across text, images and PDF content. Architecture uses Lambda orchestration, vectorized KBs and dynamic query reformulation; agents plan, decompose goals, self-reflect and update memory. Demo links two distinct KBs (university + hospital), matches volunteer students to oncology patients, drafts emails and can act via tools; system scales to batches like 100,000 documents and performs cross-model ranking and multimodal retrieval.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:03AI evolution to agent systems
- 3:50What are AI agents
- 8:03AI agent architecture components
- 10:32Agentic AI impact on RAG
- 13:16Multimodal RAG and query formation
- 16:25Agentic AI architecture recommendations
- 18:49Traditional RAG challenges and solutions
- 19:10Multidatabase system demo
- 20:53Summary
related talks
31mOrchestrating Multimodal RAG With Agentic Workflows
40mBuilding Intelligent Search Systems With Agentic RAG and OpenSearch
34mImplementing an open-source RAG with OpenSearch
01h 30mOptimizing Retrieval for Agentic Systems
27mElevating AI Applications with OpenSearch's Flow Framework and RAG Tool
39mAgentic RAG Platform for AI-Assisted Workflows
21mSmart Search with RAG: Elasticsearch Meets Language Models
37mBuilding advanced RAGs with Elasticsearch OpenSearch