Ergonomic Posture Assessment Agent

Key Highlights

• Built an AI agent that analyses human posture from images using MediaPipe Pose landmark detection (33 keypoints) and OpenCV geometric processing to assess ergonomic sitting and standing alignment.
• Developed a FastAPI REST API with endpoints for posture scoring (/analyze, 0–100 scale), natural-language ergonomic feedback (/feedback), and uptime monitoring (/health).
• Implemented a rule-based scoring engine measuring 12 angular metrics (spine curvature, shoulder asymmetry, neck tilt, hip alignment), each contributing to a weighted composite posture score.
• Containerised with Docker using multi-stage builds to minimise production image size; deployed publicly on Hugging Face Spaces with automatic health-check monitoring.
• Added a Gradio-based interactive web interface allowing users to upload images and receive annotated posture feedback directly in the browser without needing an API client.
• Optimised inference speed to under 120ms per image on CPU by caching MediaPipe model initialisation and using NumPy vectorised angle calculations instead of per-landmark loops.