A travel-planning assistant built in Foundry must answer questions about tonight's weather at a customer's destination city. The internal knowledge base has no weather data, and weather changes hour to hour. Which approach fits this requirement?
Select an answer to reveal the explanation.
Short Explanation
Weather is one of those things that is only true for a moment, so no matter how much historical data you feed a model, it can never actually know what is happening outside right now, it can only know what usually happens around this time of year. The fix is to give the assistant a way to reach out to a live source at the exact moment someone asks, so it is pulling fresh, current numbers instead of anything remembered. Training on old weather records just teaches seasonal averages, not tonight's forecast. Building a search index over last year's almanac has the same problem, it is frozen in time and cannot update itself. And showing the model an example of how to phrase a weather answer teaches it a style, not a fact, so it still has nothing real to report unless it actually goes and looks.
Full Explanation
The correct answer is C. Configuring the agent to call a real-time web or search grounding tool lets it retrieve current weather conditions at the moment the question is asked, which matches the requirement for hour-to-hour accuracy that no static internal source can provide. Option A is incorrect because fine-tuning on historical weather records teaches the model patterns from the past, such as typical seasonal conditions, but cannot report what the weather actually is tonight, since weather is not a fixed fact the model can memorize. Option B is incorrect because a retrieval-augmented generation index built over last year's almanac is also a fixed, historical snapshot; it can describe average conditions for a date but has no way to reflect what is happening right now. Option D is incorrect because a few-shot example showing typical weather phrasing only demonstrates a response format or tone, it supplies no actual data, so the model would still be guessing at real values rather than reporting them. Only a tool that fetches live data at query time satisfies a requirement that changes hourly.