FOVB-AIoT: An Automated Kiosk-Based Multi-Sensor Platform with Multi-Agent AI Consensus for Health Risk Screening

Traditional institutional health monitoring protocols, such as university Annual Physical Examinations (APEs), are heavily constrained by manual measurement, multi-station triage latency, and human recording variability. This study presents the design, deployment, and clinical evaluation of the FOVB-AIoT, an integrated kiosk platform combining automated physiological sensor acquisition, Internet of Things (IoT) telemetry, and artificial intelligence for real-time Health Risk Prediction (HRP). The platform integrates a time-of-flight LiDAR sensor, four load cells with HX711 amplification, an MLX90614 infrared radiometer, a MAX30102 photoplethysmography sensor, and an automated blood pressure unit digitized using edge-based YOLOv11 computer vision. Pre-measurement compliance is enforced via multi-view vision models verifying the absence of footwear and heavy accessories. HRP is executed through a collaborative multi-agent framework triangulating eXtreme Gradient Boosting (XGBoost), an on-device lightweight language model (GPT-OSS-120B), and a cloud-based semantic reasoning engine (Gemini 2.0) under an Expert-in-the-Loop (EITL) protocol. Empirical validation across n=72 participants confirmed high sensor accuracy (≥98.5%) against clinical reference instruments. Paired samples t-test analysis revealed statistically significant improvements over traditional triage for both time consumption (t(71)=9.33, p<0.001) and operational efficiency (t(71)=12.052, p<0.001), establishing the platform as a validated, high-throughput solution for institutional preventive healthcare.