Hands-on AI workshop: Graph RAG, Memory & Multimodal Agents
summary
Live build produces a working multimodal rescue pipeline: Cloud Spanner Graph stores entities and vectors, Gemini Pro and a text embedding model generate embeddings inside Spanner, and a hybrid RRF ranking combines keyword + semantic search as Graph RAG to match needs to helpers; the ADK orchestrates a sequential upload→extract→spanner→summarize agent chain, deployable to Cloud Run or Agent Engine, with session callbacks that persist facts into Vertex AI Memory Bank so long‑term memory loads on return. Demo shows a 3D graph, four survivors and two matched medics via the ADK Web UI.
AI-written summary · about
chapters (18)
- 0:00Introduction and setup
- 3:50Challenge overview
- 7:22Equipment and environment
- 10:21Cloud shell configuration
- 12:08Project and service setup
- 15:07Solution stack explanation
- 17:57Data preparation
- 21:16Graph visualization
- 24:05AI model integration
- 26:44Semantic search basics
- 30:44Embeddings implementation
- 37:20Building the RAG agent
- 40:00Search methodology
- 45:40Agent testing
- 49:44Full stack architecture
- 55:20Multimodal agent pipeline
- 58:26Extraction and processing
- 1:14:00Agent memory and Q&A
related talks
51mMemory Masterclass: Make Your AI Agents Remember What They Do!
30mGenAI Beyond Chat with RAG, Knowledge Graphs, and Python
29mTransparent Multi-Agent AI
01h 49mBuilding Memory Augmented Agents with MemoRizz
58mAdvanced Autonomous AI Agents: Memory, Collaboration and Evolution in Action
29mThe AI Agent Memory Landscape
37mBuilding Multimodal AI Agents From Scratch
26mStore Your AI Agent's Memory and Context in a Knowledge Graph