SMK Dharma Loka still relies on a manual attendance system that is prone to recording errors and data manipulation. This study develops an automated attendance system based on face recognition, with a case study at SMK Dharma Loka Pekanbaru. The face recognition pipeline consists of MTCNN for face detection and extraction, FaceNet Inception ResNet V1 with MS-Celeb-1M pretrained weights to generate 512-dimensional embeddings through a transfer learning approach, and an ensemble of three classifiers — Support Vector Machine (SVM), Logistic Regression (LR), and Cosine Distance — with a majority voting mechanism. The dataset consists of 496 face images from 19 students that passed MTCNN detection, enriched with 16 augmentation variations. Closed-set evaluation yielded an accuracy of 96.23% for both SVM and LR, and 95.28% for Cosine Distance. Analysis testing showed that the system performs optimally at distances of up to 3 meters under normal lighting conditions. The system is implemented on an Android application, a Flask backend, and a React-based web interface. User Acceptance Testing conducted on 5 respondents produced a score of 80.6%, indicating that the system was well received by users.