I'd like to decide which model could be used for achieve the parent goal.
A good candidate could be a model for text summarization.
Requirements would be quality and latency which would be both vague at the moment and need to be quantified.

Quality: The purpose of this task would not be to establish an evaluation framework for the underlying task (e.g. text summarization) however we need to have a way to test that it meets some standards and would be safe for use by the community

Latency: As this would be the first iteration there are no hard latency requirements but the response time would have to be something that is acceptable from a user's perspective.