We introduce a divide-and-conquer pipeline for multi-hop knowledge base question answering that decomposes relation detection into tractable sub-tasks, enabling efficient handling of multi-label supervision and improving accuracy on benchmark datasets.