MLOps Engineer

About the Role

We are looking for an MLOps Developer to own the model lifecycle, deployment pipelines, and operational health of this AI system. You will bridge the gap between model development and production, ensuring models are reliably trained, versioned, deployed, and monitored across diverse hardware environments.

Key Responsibilities

• Optimize and deploy PyTorch models (detection + video classification) to TensorRT and ONNX for real-time GPU inference across both server and edge hardware.

• Build and maintain reproducible model training and evaluation pipelines with experiment tracking and dataset versioning

• Design CI/CD workflows for model validation, packaging, and rollout to remotely deployed devices.

• Monitor production inference pipelines — GPU utilization, latency, frame throughput, and model performance drift

• Manage cloud storage (AWS S3) for model artifacts, video clips, and deployment assets

• Containerize services (Docker) and investigate Kubernetes-based orchestration for multi-site deployments for edge hardware environments

• Collaborate with ML and CV engineers to operationalize new model versions and document deployment runbooks

Requirements

• 2–4 years of MLOps or ML engineering experience in production.

• Strong Python; hands-on with PyTorch, ONNX, and TensorRT

• Experience with CI/CD tools, Docker, and Linux/bash environments

• Familiarity with experiment tracking (MLflow, ClearML, W&B, or similar)

• AWS or Azure cloud experience with exposure to ML-specific services such as:

• AWS: SageMaker (training jobs, model registry, endpoints), ECR, S3, Lambda, CloudWatch

• Azure: Azure Machine Learning (pipelines, model registry, compute clusters), Azure Container Registry, Blob Storage, Azure Monitor

Nice to Have

• NVIDIA Jetson edge device deployment experience (JetPack/aarch64)

• Real-time video processing (OpenCV, GStreamer, RTSP)

• Dataset versioning tools (DVC or similar)

• WebRTC or WebSocket-based streaming experience

• Exposure to Docker / Kubernetes for model serving at scale