In a service you need to describe data models, both internal structures and incoming JSON from external clients. When would you use a dataclass, and when a Pydantic model?

How to approach it

The interviewer checks whether the candidate understands the fundamental difference: a dataclass is code generation with no runtime checks, while Pydantic is validation and type coercion. And whether they can derive an architectural rule from that: where checks are needed, and where they're unnecessary and just cost you.