Why Prompting Isn’t Enough: Building Stateful AI Architecture

This talk teaches experienced developers how to build production-ready AI agents by treating LLMs as stateful distributed systems rather than stateless prompt-response services. It covers implementing real-time streaming architectures with Apache Kafka and Flink, creating multi-tiered memory structures to overcome token limits and cost spikes, and integrating LLMs via MCP with SQL for real-time anomaly detection workflows. Ideal for engineers building scalable, resilient AI systems.