Ask HN: What does your machine learning pipeline look like?

A typical production ML pipeline involves several stages: data ingestion and validation, orchestration, training and experiment tracking, feature store management, model serving, and monitoring with retraining. For data ingestion, common tools include Apache Beam and AWS Kinesis. Orchestration is often handled by Airflow or Kubeflow, which manage workflows and dependencies. Training and experiment tracking can be done using tools like TensorFlow, PyTorch, and MLflow or Weights & Biases for tracking experiments. Feature stores like Feast or Tecton help manage and serve features. Model serving can be achieved with TensorFlow Serving, AWS SageMaker, or Azure Machine Learning. Monitoring and retraining involve tools like Prometheus and Grafana for performance tracking, and automated retraining pipelines triggered by data drift or performance degradation.