Single-model vision pipelines are also lacking, and it means that traditional automated attendance systems cannot reliably operate in unconstrained environments. The proposed paper presents a new deep model fusion architecture, a two-stage design, to create an extremely robust and accurate face recognition system that can be used in managing real-time attendance. We propose a two-specialized hybridization approach consisting of building two hybrids: 1) a Detection Fusion Model that is based on YOLOv8 in conjunction with RetinaFace through Weighted Box Fusion (WBF) to attain higher spatial resolution (estimated WIDER FACE mAP of 96.5) and constant thinking speed (estimated 35.2 FPS), decreasing false alarms and enhancing small face recognition; and 2) a Recognition Fusion Model, which relies on the synergies between ArcFace (angular margin This combination approach creates a maximum margin between matching and non-matching identities (maximum of 0.787), which produces a high verification reliability. Using a combination of these customized fused models, the resulting system can successfully deal with multi-person recognition in group photographs, confirming an entirely correct and lowlatency solution to overcome the major drawbacks of singlemodel face recognition strategies of that time and establishing a new record of automation rates of attendance.