Building Hierarchical Multi-Agent RAG Systems
summary
A hierarchical RAG prototype uses Spring AI, Ollama plus OpenAI/Gemini models, a Python OCR container, PGVector with five to six test records, and a Salesforce sandbox to automate customer onboarding across roughly 15–20 steps. An orchestrator LLM appends growing context and dynamic orchestrator decides next agent to pick a specialist (extraction → fraud similarity check → CRM create), and the demo manages to CRM account created though the setup is brittle and some model/tool swaps were required.
AI-written summary · about
chapters (11)
- 0:00Introduction and context
- 1:27What is an agent
- 3:15Agent anatomy and components
- 5:01Single vs multi-agent systems
- 7:08Multi-agent patterns
- 9:39Demo setup overview
- 13:00Live demo walkthrough
- 21:58Orchestration and state management
- 26:11Frameworks and standards
- 28:48MCP and tool discovery
- 30:36Hierarchical RAG use cases
related talks
32mBuilding Multi-Agent RAG Systems
51mTransforming Banking with Advanced AI: GenAI, RAG, and Multi-Agent Systems
01h 16mForget RAG Pipelines—Build Production Ready Agents in 15 Mins
22mOrchestrating Multimodal RAG With Agentic Workflows With OpenSearch
43mArchitectural Blueprints for AI Agents
40mAI Agents in Practice
31mAdvanced Enterprise RAG Systems
29mTransparent Multi-Agent AI