Job Description
- Own the end-to-end computer vision roadmap — from problem framing and data strategy through model development, edge deployment, and production monitoring — across industrial portfolio (steel, cement, pharma, paints, and beyond).
- Personally architect and build solutions for the most complex vision challenges: novel defect types, extreme class imbalance, multi-camera fusion, low-light / high-noise factory environments, and real-time inference on constrained edge hardware.
- Stay at the cutting edge of CV research and rapidly evaluate and adopt new models and techniques — YOLO26, SAM 3, Vision Transformers (DINOv2, Swin), Grounding DINO, RF-DETR, zero-shot / open-vocabulary detection (YOLO-World, CLIP) — translating papers into production value.
- Define and enforce engineering standards for the vision stack: model training pipelines, data versioning (DVC), annotation workflows (CVAT, Roboflow, Label Studio), experiment tracking (W&B, MLflow), edge export formats (TensorRT, ONNX, OpenVINO), and CI/CD for model updates.
- Drive inference optimisation — quantisation (INT8 / FP16, GPTQ), pruning, knowledge distillation, and batching strategies — to meet latency and cost targets across NVIDIA Jetson, industrial PCs, and cloud GPU instances.
Required Skills
[Computer Vision, Python, OpenCV]
Additional Information
- 5–9 years of hands-on experience in computer vision — with a strong track record of taking models from research / prototyping through to production deployment.
- Deep proficiency in Python and PyTorch; strong working knowledge of OpenCV, Albumentations, and image / video processing fundamentals.
- Demonstrated expertise across multiple CV tasks: object detection, instance / semantic / panoptic segmentation, anomaly detection, pose estimation, or tracking.
- Hands-on experience with modern model families — YOLO (v8 / v11 / v26), transformer-based detectors (RT-DETR, DETR, RF-DETR), segmentation models (SAM / SAM 2), and CNN backbones (ResNet, EfficientNet, ConvNeXt, Vision Transformers).
- Production experience deploying models to edge or on-prem hardware using TensorRT, ONNX Runtime, or OpenVINO; comfort with Docker, Kubernetes, and at least one cloud platform (AWS / Azure / GCP).
- Experience in a high-growth start-up or similarly fast-paced environment where scope is ambiguous, timelines are tight, and wearing multiple hats is the norm.
- Strong first-principles problem-solving ability — comfortable navigating novel, unstructured problems where no playbook exists.
- Excellent communication skills — able to distil complex technical concepts for non-technical stakeholders, write clear documentation, and present results to leadership and customers.
