Depends what you're going for. Change detection has a lot of established use cases but not sure whether LLM interfaces will add usability or remove accuracy; object detection in Google Street View images sounds cool and novel, but not sure what the actual use case is...