We optimized performance for a real-time architectural application developed by the team behind Drawhome, focusing on reducing GPU/CPU overhead. We implemented render-on-demand to avoid unnecessary frame updates and refactored geometry generation for stairs and walls to improve buffer efficiency. We also introduced a material pooling system to reduce state changes and built a Docker-based service for KTX2 texture compression, improving memory usage and load times. Additionally, we implemented BVH-based culling for efficient spatial queries. These changes reduced draw calls by ~60%, significantly improving rendering performance.