Senior Computer Vision Engineer

ABOUT SPORTFX • SportFX is a high-performance, fully-remote startup democratizing elite sports analytics through AI—serving everyone from youth leagues to professional franchises with the same world-class technology. • We combine deep sports expertise with cutting-edge computer vision and ML to deliver real-time video analysis, personalized coaching insights, and data-driven performance optimization. • Culture: extreme ownership, radical transparency, and collaborative excellence. We operate with high-trust autonomy where smart people solve hard problems together—no egos, no politics, just results. ABOUT THE ROLE — SENIOR COMPUTER VISION ENGINEER You'll architect and own the entire computer vision pipeline that converts monocular smartphone video into precise biomechanical analysis. This is a broad technical leadership role covering 3D pose estimation, object detection/tracking, and temporal segmentation—basically everything from raw video input to structured kinematic output. Reporting to our Head of AI/CV, you'll be the technical decision-maker for our CV stack. You're not hyper-specialized in one domain—you're the engineer who can architect pose estimation in the morning, debug ball tracking at lunch, and design temporal segmentation models in the afternoon. You'll set technical direction, establish best practices, and mentor a mid-level engineer as the team scales. What You'll Own 3D Pose Estimation: Lead development of our monocular-to-3D pipeline. Work with parametric body models (SMPL-X or similar), handle occlusion and motion blur, implement temporal consistency. Get us from 95% accuracy to 99%+ through systematic improvement. Object Detection & Tracking: Build production detectors for balls, bats, clubs, rackets. Handle tiny objects moving at high speed through motion blur. Implement multi-object tracking that doesn't drop frames. Extract trajectories and equipment angles from noisy video. Temporal Segmentation: Develop frame-accurate phase detection for pitching, throwing, golf swing, tennis serve. Build models that understand biomechanical constraints—stride foot contact happens before max external rotation, not after. Multi-Sport Expansion: Lead CV development for new sports beyond baseball. Each sport has unique challenges—golf ball tracking is different from baseball, tennis serve mechanics differ from pitching. Architect solutions that scale across sports. Production Optimization: Take inference from 2+ seconds to sub-500ms through model optimization, quantization, efficient architectures, and deployment strategies. Ship models via ONNX/TensorRT that actually run fast in production. Technical Leadership: Establish CV best practices, review code and model architectures, make build-vs-buy decisions, mentor mid-level engineer. You're setting technical standards for the team as it grows. Your Environment Tech Stack: PyTorch, OpenCV, MediaPipe, SMPL-X or similar parametric models, YOLO family, ByteTrack, ActionFormer/ASFormer, ONNX, TensorRT, Docker/CUDA Pipeline: Real-time video processing, pose estimation, detection, tracking, temporal modeling, all feeding into biomechanics analysis Team: You'll work with Head of AI/CV on strategy, mentor Mid-Level CV Engineer on execution, collaborate closely with OpenSim/Biomechanics Engineer on validation, coordinate with DevOps on deployment Workflow: Rapid experimentation with W&B tracking, validation against motion capture ground truth, bi-weekly accuracy reviews with quantitative metrics Location: Global (4+ hour overlap with US Central Time required) SKILLS & EXPERIENCE Must-Have • 5+ years production CV/ML experience shipping models that serve real users at scale • Deep PyTorch expertise including custom architectures, training optimization, and debugging complex systems • Strong 3D computer vision fundamentals—multi-view geometry, camera calibration, 2D-to-3D methods, depth estimation • Proven experience with pose estimation systems (MediaPipe, OpenPose, ViTPose, or similar) • Object detection and tracking production experience—you've built systems that work on real video, not just benchmark datasets • Video processing expertise including temporal modeling, handling variable frame rates, motion blur, occlusion • Production deployment skills with model optimization (ONNX, TensorRT, quantization) • Technical leadership—you've mentored engineers, made architectural decisions, established best practices Nice-to-Have • Experience with parametric body models (SMPL, SMPL-X) • Background in temporal action segmentation or sequence modeling • Published research in computer vision (CVPR, ICCV, ECCV) • Understanding of biomechanics or sports science • Multi-object tracking in challenging outdoor conditions • Mobile CV deployment (TensorFlow Lite, Core ML) • Experience with synthetic data generation • Domain expertise in sports (baseball, golf, tennis, football) What Sets You Apart: You're a generalist who goes deep when needed. You don't just specialize in one narrow CV domain—you understand the full pipeline and can architect end-to-end solutions. You've shipped CV systems in production and know the difference between 99% accuracy in the lab and 95% accuracy on shaky phone video captured at a youth baseball game. You can context-switch between pose estimation, detection, and temporal modeling without losing velocity. Job Types: Full-time, Part-time, Contract Pay: $75,000.00 - $160,000.00 per year Benefits: • Paid time off Application Question(s): • Describe an end-to-end computer vision system you architected — from raw video input to final outputs. Explain the major components (detection, tracking, pose, temporal modeling, post-processing), how they interacted, and one major bottleneck you had to solve. • You get noisy, jittery 3D keypoints from a monocular pose lift model. Walk us through how you would: 1. Diagnose the source of the noise 2. Improve temporal consistency 3. Handle occlusions and dropped joints 4. Validate improvements quantitatively • Explain a time you worked on detecting or tracking small, fast-moving objects (balls, bats, equipment, drones, vehicles, etc.). Describe the failure modes you encountered at high speed or under motion blur, and the specific architectural or augmentation strategies you used to fix them. • Share links to 2–4 of your strongest computer vision projects (GitHub, repositories, demos, research papers, videos). For each, briefly explain your role, the core technical challenges, and what impact your solution had in production. Work Location: Remote Apply tot his job