Senior ML / MLOps Engineer - Long-Term Engagement

We're looking for a senior ML / MLOps engineer. Long-term engagement, starting part-time with strong potential to grow into full-time. This is not a pure infrastructure role: you'll own both the ML infrastructure and the models themselves. You should move comfortably between a model's confusion matrix and its Kubernetes deployment manifest. WHAT WE EXPECT - Design and run model drift monitoring in production (data/concept drift, alerting, dashboards) - Build automated evaluation and performance gates into CI/CD pipelines for model promotion - Implement and enforce a Model Registry workflow (MLflow) with approvals and full lineage - Deploy and operate model serving on Kubernetes (ONNX Runtime, KServe or equivalent), including canary rollouts and rollback - Understand, evaluate, and improve existing models: rigorous evaluation (golden datasets, per-class performance, failure modes), retraining and model improvements when needed - Deploy CV models on-device (mobile) for real-time auto-detection during camera capture: model conversion and quantization (ONNX Runtime Mobile / LiteRT / Core ML), working with camera frames and device sensors - Work in a disciplined, well-documented way: design notes, reproducible pipelines, traceable changes REQUIREMENTS - 5+ years in ML engineering / MLOps with production Kubernetes (EKS preferred) - Hands-on model development (PyTorch/TensorFlow) and computer vision experience - MLflow (tracking + registry), model serving (ONNX Runtime / KServe / Triton) - GitHub Actions or equivalent, Terraform or comparable IaC, Python - ML observability: Prometheus/Grafana, drift detection approaches - Clear written English Nice to have: on-device / edge ML deployment (LiteRT, Core ML, ONNX Runtime Mobile) with real-time camera-frame inference; AWS (EKS, Step Functions, SageMaker); Flagger/Argo Rollouts; classical image processing. ENGAGEMENT Long-term (6+ months, likely much longer). Async-first via Slack/GitHub, weekly syncs; 2-3 hours of overlap with UTC+3 needed. NDA required. HOW TO APPLY In your proposal, tell us about the most relevant production ML system you've owned and why this role fits you. Start your proposal with the word "PYTORCH" so we know you've read the full posting. Please also answer the screening questions - proposals with skipped or generic screening answers will not be considered.