Mobile & Computer Vision Engineer: Refactor MediaPipe Real-Time Pose Tracking

We have a React Native mobile application that uses MediaPipe BlazePose keypoint estimation to track real-time human movement sequences and evaluate repetition counts for various physical movement patterns. Currently, our tracking engine suffers from coordinate jitter, frame-drop latency, and inaccurate state tracking (e.g., false-positive counts, partial movements triggering completion, or double-counting). We need an experienced Mobile / Computer Vision engineer to refactor our pose evaluation pipeline so tracking is rock-solid, smooth, and predictable. What You Will Be Working On: Keypoint Smoothing & Filtering: Implement spatial/temporal filtering (such as Exponential Moving Average or Low-Pass filtering) on raw landmark coordinates to eliminate noise before calculating joint vectors and angles. Hardened State Machine Architecture: Re-architect our movement evaluation using strict dual-threshold hysteresis (distinct entry/exit bounds), state duration guards, and landmark confidence filtering. Offline Testing Setup: Set up a clean local script/test harness using pre-recorded video feeds (clean movements, fast movements, partial movements) to log and verify state transitions frame-by-frame before running on-device. Required Experience: Strong proficiency in React Native (iOS/Android) and real-time camera/WebView performance optimization. Production experience with MediaPipe (BlazePose), OpenCV, or similar pose-estimation pipelines. Firm grasp of 3D vector geometry and Finite State Machine (FSM) software patterns. To Apply: Please describe your experience working with MediaPipe or pose-estimation frameworks, and briefly explain how you typically handle landmark jitter and false state triggers in real-time mobile applications.