We need a new ContentRetriever implementation that can answer natural language questions by querying a…

We need a new ContentRetriever implementation that can answer natural language questions by querying a…: a task in LegoFlow-SWE (Harbor dataset). The serialised output must faithfully represent the data that was actually fetched.