Event-Driven & Multi-Agent Architectures for Gen AI Workflows
summary
Argues that combining event sourcing + CQRS with multi‑agent orchestration turns one‑shot LLM calls into reproducible, scalable workflows; demonstrates using streaming brokers (examples: Apache Kafka, Pulsar, Redpanda) to capture state, replay orders, and back RAG pipelines that call vector DBs and LLMs; shows small reference implementations and SDKs (autogen / CrewAI / LangGraph / Strands) and notes practical tradeoffs: harder debugging, observability needs, hallucinations, and high compute cost.
AI-written summary · about
chapters (17)
- 0:00Introduction and agenda
- 1:05Speaker background and community work
- 3:35Demystifying GenAI and agentic architecture
- 6:44Why agents are essential for LLMs
- 8:30Agentic architecture key concepts
- 10:58Event-driven world and AI data states
- 13:04Event sourcing and CQRS explained
- 15:33GenAI streaming usages and RAG
- 17:18Benefits of event-driven approach
- 18:41Challenges of event-driven systems
- 19:24CAP theorem in distributed computing
- 24:01GenAI apps and CAP theorem trade-offs
- 24:22Retrieval augmented generation deep dive
- 27:07Applying event-driven to agents and workflows
- 28:53Multi-agent libraries and tools
- 29:35Challenges in GenAI applications
- 30:38Closing and contact information
related talks
50mHarnessing Event-Driven and Multi-Agent Architectures for Complex Workflows in Gen...
56mDesign Patterns for Multi Agent Systems
31mAgentic AI architecture patterns
44mBuilding Agentic AI Systems: From Transactions To Conversations
56mHerding LLMs: Predictable Outcomes from Unpredictable Agents
29mAgentic AI Meets Event-Driven Architecture
43mArchitecting the Future: Modern Solution Design with Agentic AI
50mScalable Innovation: Exploring Multi-Agent Systems