ML Infrastructure Engineer

Role overview Build and operate the infrastructure that trains, versions, deploys, and serves multiple computer vision and fraud detection models. The work spans cloud systems, model delivery workflows, inference performance, and operational monitoring, with a stated target of keeping API responses under 200 ms while supporting auditable changes and zero-downtime releases.

Responsibilities - Develop and maintain model-serving infrastructure on Google Cloud, using services such as Cloud Run, GKE, or Vertex AI. - Create automated pipelines for model training, evaluation, approval, and production rollout. - Introduce model versioning, canary releases, and A/B testing. - Reduce inference latency through batching, quantization, and hardware-aware compilation. - Monitor model quality, drift, and request throughput, and configure alerts. - Allocate GPU and TPU capacity across training and serving, balancing performance with cost.

Requirements - At least three years in ML infrastructure, MLOps, or backend systems engineering. - Strong Kubernetes and Docker experience, plus cloud deployment work on GCP or AWS. - Practical experience with a model-serving framework such as Triton, TorchServe, or TensorFlow Serving. - Python proficiency and working knowledge of Go, Rust, or C++. - Experience building ML delivery pipelines with tools such as MLflow or Kubeflow, or custom equivalents. - Understanding of GPU inference optimization, including TensorRT or ONNX Runtime.

Nice to have - Production experience with Vertex AI or SageMaker. - Work on real-time services with strict latency targets, especially below 200 ms at p99. - GPU workload cost estimation, distributed multi-GPU or multi-node training, or infrastructure-as-code with Terraform or Pulumi.