Processing planning applications is a time-intensive task for local governments, involving diverse documents such as technical drawings, location plans, and written statements. To streamline this workflow, we present a lightweight, two-stage classification pipeline for automatically identifying document types within planning application submissions. In the first stage, an XGBoost classifier distinguishes between text-based and drawing-based pages using pixel-density histograms. A second stage refines this classification by using textual features to separate technical diagrams from non-technical illustrations, addressing cases where visual features alone are ambiguous. The full system achieves 99% document-level classification accuracy, offering a scalable and reliable solution suitable for deployment in real-world public-sector environments.