In this project, we'll build three end-to-end, database-backed features that consume structured LLM output for a hotel booking application. In this type of application, an LLM response can drive downstream actions rather than being displayed directly, such as looking up matching hotels, updating a guest rating, or opening a support ticket. We'll validate structured output against Pydantic schemas so the rest of the application can verify that the response matches the expected schema before using it, creating a clear flow from model output to validated application data and downstream logic.
We'll start by defining the database models and Pydantic schemas shared across the features. We'll then build each feature's pipeline function step by step: validating the incoming request, calling the LLM with a system prompt that specifies the expected fields and output format, parsing and validating the response against a Pydantic schema, and persisting the validated result to the database. Using hotel search, guest review analysis, and support chat ticketing scenarios, we'll turn a plain-language query into matched hotel results, a raw guest comment into structured sentiment analysis, a recommended rating, and a drafted reply, and a guest message into a routed, prioritized support ticket with an automatic acknowledgment. By the end of this project, we'll have hands-on experience integrating LLM calls into a reusable backend workflow with validation, persistence, and error boundaries, along with a clear understanding of how structured model output flows into validated application logic.