フリー問題

問題文

A nightly pipeline loads raw product reviews and must add two derived columns: a category label and a one-line summary. The team wants this to be part of the existing SQL transformation, not a separate service. Which approach fits?

選択肢

1. Export the reviews to a stage, call the REST API from an external scheduler, and load the results back, since AI functions cannot be used in statements that write to a table.
2. Put the logic in a container service where a long-running process reads the table and writes the derived columns back.
3. Call the AI functions inside the transformation statement so the derived columns are produced in the same pass that writes the target table.
4. Create a fine-tuned model for each derived column, so the sentiment, the topic, and the summary each come from a model trained on that one task.