machine learning engineer for workplace safety

Lumenalta delivers consulting, engineering, strategy, software development, and custom IT system services.

• Design and build detection pipelines for client camera footage, including person and object detection, multi-object tracking, pose-based region derivation, and classifiers with a “cannot determine” outcome;
• Extend pipelines with instance segmentation, temporal reasoning, and zone-aware logic;
• Lead labelled dataset construction, including labelling guidelines, adjudication, class imbalance handling, video split hygiene, and hard case mining;
• Train, fine-tune, and calibrate models against contractual performance targets;
• Own held-out evaluation methodology, including sample selection, adjudication, and reporting of TP, FP, TN, FN, error rates, and confidence intervals;
• Derive person-level and event-level outputs with duration for operational dashboards and reviewer alerting;
• Build systems that generalise across camera viewpoints and sites;
• Architect inference so deployment location can change between cloud batch and edge inference;
• Own frame sampling, motion and zone gating, CPU/GPU allocation, and batch-versus-streaming trade-offs;
• Manage model versioning, drift detection, and retraining based on human reviewer decisions;
• Write camera placement, positioning, and image quality recommendations;
• Deliver recorded enablement sessions and work alongside client-nominated technical staff.

• 5+ Years of experience in applied machine learning;
• At least 3 years of computer vision experience shipped to production and operating on real-world footage;
• Deep hands-on experience with object detection and multi-object tracking;
• Experience with YOLO, Faster R-CNN, RT-DETR or RF-DETR, and SORT or DeepSORT;
• Experience with instance segmentation and converting pixel measurements into real-world units through camera calibration;
• Experience with human pose estimation or keypoint-based region localisation;
• Experience building labelled datasets from raw footage, including labelling strategy, severe class imbalance handling, and prevention of video split leakage;
• Knowledge of precision and recall trade-offs, operating point selection, and probability calibration;
• Practical experience handling domain shift across cameras and locations;
• Working knowledge of distributed processing with Databricks and Spark;
• Proficiency in Python and PyTorch or equivalent;
• Awareness of open-source licensing implications for commercially delivered models, including copyleft terms;
• Experience working under data governance constraints, including client-tenant-only access and restrictions on local footage copies;
• Strong written and spoken English;
• Confidence presenting methodology and defending results to client stakeholders;
• Nice to have: PPE detection, workplace safety, video surveillance analytics, temporal action recognition, behaviour classification over video, multimodal work with sensor or time-series data, fine-grained discrimination between visually similar classes, edge inference deployment such as NVIDIA Jetson, Databricks ML tooling including MLflow and Unity Catalog, human-in-the-loop review systems, animal or livestock monitoring, consulting or client-facing delivery experience.

• Fully remote;
• Must have availability to work overlapping U.S. Pacific, Central, or Eastern time zones.