The charter desk at Meridian Ocean Carriers wants a tool that answers clause-level questions about charterparty contracts and always cites the clause number it drew from. Which model task should the GenAI engineer design the pipeline around?
Select an answer to reveal the explanation.
Short Explanation
You need a pipeline that can find the exact clause, pass it to the model with the question, and answer only from that retrieved text. That's retrieval-augmented question answering over indexed contract clauses, because the desk wants clause-level answers with citations.
Full Explanation
Retrieval-augmented question answering over indexed contract clauses is correct because the requirement has two hard constraints that only retrieval-grounded generation satisfies: the answer must be scoped to a specific clause, and it must be traceable back to that clause. Text summarization of the full charter contract into a one-paragraph digest is wrong because it condenses the whole document rather than answering a targeted question, and it discards the citation trail the desk explicitly needs. Zero-shot text classification of contracts into risk tiers is wrong because it produces a category label (a risk tier), not a natural-language answer to an open-ended clause question, and there is no fixed label set for arbitrary charter questions. Machine translation of the contract into the charterer's local language is wrong because nothing in the ask suggests the charterer reads a different language; introducing translation solves a problem the desk never raised. The distinguishing signal in a 'select the model task' item is matching the verb in the business ask (answer, with citation) to the task that structurally produces that output, not picking whichever task sounds generically appropriate for a document-AI project.