Bringing stories to life with AI
summary
Claim: generative agents actually evolve in real time by wiring LLMs into streaming infrastructure. Concrete stack: Apache Kafka + Flink pipeline streams 20–30 character locations per second, Flink emits proximity events, LLMs generate utterances, conversations and summaries feed a vector store so RAG with vector search recreates long‑term memory. Built a Stardew Valley village that spreads a goal (one NPC’s conference) across agents; it runs but needs cooldowns, aborts and prompt tuning to stop chatter and memory drift.
AI-written summary · about
chapters (10)
- 0:00AI and streaming data
- 1:53LLM agent research papers
- 10:20Interactive AI games with agents
- 19:38Stardew Valley villagers
- 21:44Kafka for real-time movement
- 24:34Flink for proximity
- 27:04AI-driven conversations
- 29:36Long-term memory with RAG
- 33:35Villager goals and ideas
- 36:43Summary and resources
related talks
41mBringing stories to life with AI, data streaming and generative agents
51mEvent Driven AI: Supercharging ChatGPT with RAG & LangStream
27mChallenges of Ambient Agents: Building Governable AI Systems at Scale
44mBuilding Agentic AI Systems: From Transactions To Conversations
16mTransform customer experience with an array of AI agents
48mGoing Beyond Prompt Engineering
46mDemystifying AI Agents
01h 04mBuilding Expertise Into Generated Conversations