AI Engineer - Computer Vision

Responsibilities:

• The core on-device CV stack detection, tracking and per-feature analytics engineered to run reliably within the compute and memory budget of edge hardware, at production quality across a growing device fleet.
• Model performance on the edge: taking models built for the GPU and making them fast and accurate on constrained devices: NCNN/ONNX export, int8 quantisation, resolution/cadence tuning, and a shared-inference design that lets one box serve many features.
• The analytics feature set: footfall & demographics, dwell and zone/table occupancy, queue and service-time, PPE/SOP compliance, and cross-camera person re-identification.
• Accuracy in the real world: building the evaluation harnesses and ground-truth workflows that hold every feature to a measurable bar across varied camera angles, lighting and occlusion, rather than trusting benchmark FPS.
• The path from model to fleet, partnering with the platform team on deployment, OTA model updates and monitoring so accuracy holds across every store, not just the lab. A voice in the CV roadmap and engineering standards: which models we invest in, how we evaluate, how we ship and helping level up the vision engineers around you.

Requirements:

• 3+ years of computer-vision / deep-learning engineering with systems that reached production.
• Deep hands-on with object detection (YOLO family) and multi-object tracking (ByteTrack, DeepSORT or similar) on real camera / RTSP video.
• Real experience optimising and deploying models for inference ONNX, quantisation (int8), and at least one edge/runtime stack (NCNN, TensorRT, OpenVINO or TFLite) on devices like Raspberry Pi or Jetson.
• Strong Python and OpenCV, and a genuine feel for the accuracy-latency-cost trade-off with the rigour to measure it.
• Ownership instinct: you can take a fuzzy problem to a shipped, monitored feature and hold yourself to production standards.

Strong pluses:

• Person re-ID, pose, fine-grained age/gender, or vision-language models for scene understanding.
• Edge accelerators (Hailo, Coral, Jetson), Linux/systemd, containers, and an MLOps mindset (CI, model registry, OTA).
• Retail, surveillance or smart-camera analytics domain experience.
• Building annotation pipelines and evaluation frameworks.