Event Streaming for Multi-Agentic Generative AI Workflows
summary
Argues that event streaming can coordinate multi-agent generative workflows by piping real-time inputs through a streaming layer (example: capture tagged tweets into a Kafka topic, build prompts and push them to an LLM) and by using RAG with a live vector store. Recommends specific tooling: Kafka/Pulsar, RAG + Vector DB (PGVector, Milvus, Chroma, Qdrant) and agent frameworks crewAI, LangGraph, Microsoft Autogen. Presents prototypes (RisingWave stream processing) as feasible but exploratory, noting hallucination, observability and cost challenges.
AI-written summary · about
chapters (11)
- 0:00Intro
- 2:00Agenda
- 4:14A brief intro & background of GenAl
- 14:35Understanding the "players" in the GenAI era
- 16:35The event-driven world & AI
- 27:19Retrieval-Augmented Generation (RAG)
- 29:26Vector DB & similarity search
- 34:49Agents & multi-agentic workflows
- 44:04Challenges of GenAI
- 44:35Resources
- 45:43Outro
related talks
46mEvent Streaming and its Potential for the GenAI World
42mAgentic AI with AWS Streaming
41mAdding Generative AI to Real Time Streaming Pipelines
01h 47mRadanalytics.io streaming applications workshop
50mThe tech behind modern AI/ML use cases in Gaming
58mAI-Enhanced Big Data: Integrating Private LLMs and Vector Databases
53mGenerative AI: Overview and use cases
26mEngineering Excellence: Platforms and Architectures for Real-Time AI Analytics