A bank wants to flag emails that contain the word mortgage so a queue can review them. No other judgment is required. What should the design use?
Select an answer to reveal the explanation.
Short Explanation
You do not build a laboratory to notice one word on a page. You use a keyword check or a language tool that finds mortgage, and you do not design a custom Foundry model for it. Training a model on mailbox exports is a costly way to do a find.
Full Explanation
A custom model is for behavior that prompting, grounding, and ordinary tools cannot produce. Detecting the word mortgage in an email is a closed text match, so a keyword check or a language tool is enough and a custom Foundry model is not the design. A custom Foundry model trained on mailbox exports so it can recognize that single word spends a training project on a match a finder already does, and it will drift when the wording is exact and the model is not. A fine-tuned model whose only target is to mark messages that include the word mortgage is the same custom-model project under another name. Computer Use that opens each mailbox message on screen and looks for the word by eye drives a user interface for a search the mail text already supports, which is slower and easier to break than reading the message body. Exam caveat: if the bank later needs a judgment, such as whether a message is a mortgage application rather than a mention of the word, the design can be revisited. This request stops at the word. Operational check: confirm the pipeline flags a test message that contains the word and does not open a custom-model training job to do it.