Discover how a domain-specific language (DSL) in Scala can help you design secure and composable prompts for LLMs. This talk explores techniques to prevent prompt injection, enforce prompt structure, and reason about non-deterministic behavior using a monadic effect system and Functional Programming. Also, we will show how to use automatic serialization libraries (like Circe, Play or ZIO Schema) and Scala case classes to model LLM structured responses in JSON format.