Smart Visitor Analytic (SVA) is a GPU-accelerated edge AI platform for real-time detection, tracking, and analytics. I built a Python pipeline using YOLOX Tiny, ByteTrack, and OSNet to detect people, maintain tracks, and identify likely returning visitors through appearance-based re-identification. The system uses ONNX Runtime, PyTorch, and CUDA acceleration with CPU fallback, stores campaign and visitor data in PostgreSQL, exposes monitoring and diagnostic endpoints through FastAPI, and provides an MJPEG browser dashboard with occupancy, visitor counts, camera status, and inference FPS.