The accounts of the continuing Palestinian Nakba encompass considerable significance. Over the course of the three years of the conflict, millions of photos from social media have been preserved. The preservation and classification of these data through artificial intelligence tools are essential to guarantee their availability, accessibility, and applicability. This paper presents a highly optimized, resource-constrained machine learning pipeline for binary image classification. The system is designed for the NakbaArchiveClassifier Shared Task 2026, which aims to distinguish between destroyed infrastructural images and intact infrastructural images. The system depends on two lightweight EfficientNetB0 networks to build a weighted ensemble system. Using strict hardware limitations of 2GB GPU VRAM, the system achieves an F1-score of 84.16%, which ranked 9th on the leaderboard.