Search-and-rescue (SAR) operations often occur in environments where visibility is limited, time is critical, and traditional detection methods are either unreliable or cost-prohibitive. This study presents WiFind, a context-aware pervasive system that deploys low-cost Wi-Fi nodes to detect, localize, and estimate the body pose of trapped individuals through walls, debris, and smoke. Unlike thermal imaging or wearable sensor systems, WiFind operates entirely through device-free wireless sensing, requiring no line-of-sight (LoS) or prior instrumentation of the subject. The system integrates continuous channel state information (CSI) acquisition, advanced signal denoising, and a transformer-based deep-learning architecture to infer skeletal joint positions with sub-meter localization accuracy. Experiments in both controlled and simulated SAR environments demonstrate detection accuracy above 94%, mean pose estimation error of 14 cm, and inference latency below 700 ms. Costing under $150 per node and performing robustly in high-interference, low-visibility conditions, WiFind shows the potential of pervasive wireless sensing for real-time, non-invasive survivor detection, including deployment via drones or aerial platforms for rapid coverage in inaccessible areas.