Real-time attendance automation requires high recognition accuracy, low latency, and scalable deployment, which remain challenging for conventional cloud-centric or disk-based biometric systems. This study presents NeuraAttend, a hybrid edge–cloud artificial intelligence framework designed for real-time multi-face attendance using embedding-based recognition and in-memory data management. The system integrates the Buffalo SC face detector with ArcFace-based 512-dimensional embeddings to generate discriminative identity representations robust to pose, illumination, and expression variations. To minimize end-to-end response time, similarity matching and attendance logging are performed using a Redis in-memory database, enabling microsecond-level data retrieval and sub-second logging performance. Inference acceleration through ONNX Runtime allows efficient deployment on resource-constrained edge devices while maintaining cloud synchronization for scalability and centralized management. The framework was evaluated using a custom real-world enrollment dataset collected under operational classroom and office conditions. Experimental results demonstrate a recognition accuracy of 98.6% with an average processing latency of approximately 180 ms per recognition event. Error rate analysis further confirms low false acceptance and controlled false rejection at the operational similarity threshold. By jointly optimizing model inference, storage architecture, and deployment topology, the proposed system achieves reliable, scalable, and privacy-conscious attendance automation suitable for institutional environments.