Edge-AI Wearable Smart Health Monitoring

Wearable patient monitoring on CPU hardware produced latency above 800 ms, too slow for real-time clinical alerting.

I deployed FP16 and quantized INT8 CNN models on a Jetson Orin Nano via CUDA, reaching 97–98% classification accuracy and under 100 ms end-to-end inference — a 4×+ speedup over a Raspberry Pi CPU baseline.

A wearable IoT sensor array feeds a real-time inference pipeline that raises automated anomaly alerts and supports continuous remote monitoring without cloud dependency.

I handled the full lifecycle: dataset curation, training, CUDA optimisation, hardware flashing, benchmarking and documentation.