Now wrap the model in real software. You will direct your coding agent to scaffold a FastAPI app with a `POST /ask` endpoint that takes a question, calls the model, and returns the answer plus the tokens used.

FastAPI matters here because it gives you typed request and response models, which is the first line of reliability: the shape of what goes in and out is defined and validated, not hoped for. You read the generated code and understand each part, even though the agent wrote it.