Job
Description
About the Role: We are seeking an experienced MLOps Engineer with a strong background in NVIDIA GPU-based containerization and scalable ML infrastructure ( Contractual - Assignment Basis) . You will work closely with data scientists, ML engineers, and DevOps teams to build, deploy, and maintain robust, high-performance machine learning pipelines using NVIDIA NGC containers, Docker, Kubernetes , and modern MLOps practices. Key Responsibilities: Design, develop, and maintain end-to-end MLOps pipelines for training, validation, deployment, and monitoring of ML models. Implement GPU-accelerated workflows using NVIDIA NGC containers, CUDA, and RAPIDS . Containerize ML workloads using Docker and deploy on Kubernetes (preferably with GPU support like NVIDIA device plugin for K8s) . Integrate model versioning, reproducibility, CI/CD, and automated model retraining using tools like MLflow, DVC, Kubeflow, or similar . Optimize model deployment for inference on NVIDIA hardware using TensorRT, Triton Inference Server , or ONNX Runtime-GPU . Manage cloud/on-prem GPU infrastructure and monitor resource utilization and model performance in production. Collaborate with data scientists to transition models from research to production-ready pipelines. Required Skills: Proficiency in Python and ML libraries (e.g., TensorFlow, PyTorch, Scikit-learn). Strong experience with Docker , Kubernetes , and NVIDIA GPU containerization (NGC, nvidia-docker) . Familiarity with NVIDIA Triton Inference Server , TensorRT , and CUDA . Experience with CI/CD for ML (GitHub Actions, GitLab CI, Jenkins, etc.). Deep understanding of ML lifecycle management , monitoring, and retraining. Experience working with cloud platforms (AWS/GCP/Azure) or on-prem GPU clusters. Preferred Qualifications: Experience with Kubeflow , Seldon Core , or similar orchestration tools. Exposure to Airflow , MLflow , Weights & Biases , or DVC . Knowledge of NVIDIA RAPIDS and distributed GPU workloads. MLOps certifications or NVIDIA Deep Learning Institute training (preferred but not mandatory). Show more Show less
