Posebrina

Engineered a real-time posture detection application achieving 87% accuracy by integrating MediaPipe (BlazePose) for spatial feature extraction with a lightweight MLP Classifier.

Curated a custom dataset of 1,300+ structural pose keypoints and deployed the model to provide continuous, automated posture monitoring with a 30-minute warning threshold.

Implemented asynchronous, low-latency video streaming using Streamlit-WebRTC, enabling seamless client-to-server AI inference without heavy GPU dependencies.