This study aimed to evaluate the effectiveness of deep learning-based geometry instruction using miniature roof construction in improving students’ understanding of angles. A quasi-experimental design was conducted at SMPN 1 Kotabaru with 64 seventh-grade students divided into experimental and control groups. The experimental group participated in eight sessions of project-based learning that integrated contextual modeling and collaborative exploration, while the control group received conventional textbook-based instruction. Students’ comprehension of angle concepts including classification, measurement, and application was assessed using a validated geometry test and structured reflection journals. The results showed that students in the experimental group demonstrated significantly greater improvement in angle understanding compared to the control group. Statistical analysis confirmed the effectiveness of the intervention, with a large performance gap favoring the experimental group. These findings suggest that deep learning strategies, when combined with hands-on modeling and contextual relevance, can substantially enhance conceptual mastery in geometry education.