In this session, we will explore TypeChat, an new library designed to transform the approach to building natural language interfaces using Large Language Models (LLMs). In contrast to conventional prompt engineering, TypeChat streamlines the construction of natural language interfaces through a technique referred to as „schema engineering“. We will begin with an overview of the challenges inherent to traditional decision tree-based methods and prompt engineering techniques. Following this, we will introduce the concept of „types“, the foundation of TypeChat’s schema engineering. We will delve into how developers can define types to encapsulate various user intents in their applications. The session will explore TypeChat’s automated prompt construction, demonstrating how, once types are defined, TypeChat can automatically generate the appropriate prompt for the LLM. We will also explore TypeChat’s response validation feature, showing how it ensures the LLM’s output adheres to the predefined schema and repairs the output if necessary.