About Us We combine computer-vision assessments, wearable integrations, and personalized content to help people move better. We’ve already shipped real-time overhead squat analysis to thousands of users across our mobile Android & iOS and our Enterprise iPad apps. Next up: an on-device Single-Leg Balance test that quantifies sway, stability, and compensations within seconds. The Challenge You’ll Own Lead the end-to-end R&D, model training, and production deployment of our single-leg balance assessment pipeline—from raw video capture to interpretable scores surfaced in SwiftUI & Kotlin. Research & Metric Design Define clinically relevant KPIs (center-of-mass sway, time-to-failure, hip strategy classification, knee stability, etc.). Translate research papers and PT standards (e.g., Y-Balance, CTSIB) into algorithmic requirements. Data Pipeline Drive collection strategy (camera angles, frame rate, subject diversity) and labeling guidelines. Set up annotation workflows in CVAT or Labelbox. Automate quality control and inter-rater reliability checks. Modeling & Signal Processing Evaluate pose-estimation backbones such as Google ML Kit Pose, MediaPipe, OpenPose, and custom Core ML models. Prototype sway-trajectory filters, spectral analysis, and compensation heuristics using Python and NumPy. Train and benchmark lightweight models suitable for on-device inference on iOS (A12–A17) and Android (Snapdragon 845+). Edge Deployment Convert models to Core ML or TensorFlow Lite with quantization and accelerator delegates (Neural Engine, GPU, Hexagon DSP). Integrate models into the existing Assessment Engine SDK. Expose clean Swift and Java wrappers and sample unit tests. Accuracy & Performance Validation Establish gold-standard benchmarks using force-plate or motion-capture ground truth. Meet target performance: latency under 100 ms/frame, minimal battery drain, and model footprint under 30 MB. Collaboration & Documentation Publish technical specifications, experiment logs, and API documentation for frontend, product, and external teams. Collaborate closely with iOS, Android, and backend engineers. Tools & Technologies You’ll Work With R&D: Python, Jupyter, NumPy, PyTorch, TensorFlow OpenCV, SciPy, pandas AWS Sagemaker or custom GPU rigs CVAT, Labelbox, Weights & Biases Production: Core ML, ML Kit, MediaPipe Swift, Kotlin C++/Metal (for micro-optimizations) Assessment Engine SDK GitHub Actions, Fastlane, Bitrise Must-Have Qualifications 4+ years building and deploying CV/ML models for pose, action recognition, or biomechanics. Shipped at least one model on-device (Core ML, TFLite, NNAPI, or similar) with strict latency and size requirements. Strong foundation in linear algebra and signal processing; capable of implementing filters and kinematic metrics. Experience designing data-collection protocols and managing annotation quality. Solid software engineering practices: version control, code reviews, automated tests, and clear English documentation. Nice-to-Haves Experience in rehab, sports science, or digital therapeutics. Familiarity with Rook wearable SDK, HealthKit, or aligning CV metrics with physiological data. Knowledge of 3D pose estimation (dual-cam or monocular) and sensor fusion techniques. Contributions to open-source computer vision or machine learning frameworks.