Computer Vision Engineer (Sports Video Analytics)

Overview

We’re building next-generation sports analytics for rugby — using video AI to detect, classify, and track game events like scrums, lineouts, and attacking phases. You’ll lead the computer-vision side of our stack: developing event-detection models, training feature-extraction networks, and supporting real-time inference on the cloud. This is a hands-on,role suited to someone who enjoys shipping models that run at production scale and is excited about learning the full CV development and deployment stack.

Key Responsibilities

• Develop and train models for rugby event and phase detection (e.g. scrums, lineouts, open-play transitions).
• Design and implement pipelines for data curation, augmentation and management.
• Build and optimise embeddings for player tracking and team identification.
• Implement temporal action localization and weakly-supervised learning techniques.
• Evaluate model accuracy and latency trade-offs; export to ONNX / TensorRT for real-time inference on GCP.
• Collaborate with the MLOps/Backend engineer to define the input–output schemas for APIs and ensure end-to-end deployment readiness.

Required Skills

• Languages: Python (core), basic SQL.
• Frameworks: PyTorch (preferred), TorchVision, PyTorch Lightning.
• Libraries / Tools: NumPy, pandas, scikit-learn, OpenCV, FFmpeg, Matplotlib, Seaborn, Plotly, FiftyOne.
• Deployment Tooling: ONNX Runtime.
• Video Dataset Management:
  • Proven experience working with large-scale video datasets (hundreds to thousands of clips) for training and validatio.
  • Skilled in dataset curation, cleaning, and augmentation — balancing, filtering, temporal sampling, and clip segmentation
• Version Control: Git / GitHub; experience with experiment tracking (Weights & Biases, MLflow).
• Strong understanding of modern computer-vision and video-understanding techniques.
• Familiarity with object detection, Re-ID, and tracking pipelines (e.g. DeepSORT, ByteTrack).

Nice-to-Have

• Background or strong interest in rugby / team sports analytics.
• Experience with weakly-supervised or self-supervised representation learning.
• Practical exposure to multi-camera or multi-view datasets (e.g., broadcast + tactical angles in sports)
• Experience using or integrating annotation and dataset-management tools, such as:
  • CVAT, Label Studio, Supervisely, V7, or FiftyOne for visualization and active learning loops
  • Familiarity with Roboflow, Weights & Biases Artifacts, or ClearML for dataset versioning and experiment tracking

Profile

• Seniority: Junior-Mid (3–6 years experience).
• Engagement: Full-time or 4–5 days/week.
• Location: Remote (South Africa).