Building impressive AI demos is easy - building AI systems that work reliably in production is hard. This talk dives into battle-tested patterns and techniques for creating robust RAG (Retrieval-Augmented Generation) applications and agentic workflows that your users can depend on. We'll cover everything from handling hallucinations and managing context windows, to implementing proper error handling, semantic routing and monitoring for AI systems. Through real-world examples and hard-learned lessons, you'll discover practical strategies for moving beyond proof-of-concept to production-ready AI applications.