MLflow is an open-source platform for managing the entire life cycle of machine learning models, originally introduced by Databricks in 2018. It records experiments with their parameters and metrics, manages model versions in a registry and simplifies deployment. More recent versions also support tracing and evaluating applications built on language models. Teams can therefore trace which model was produced with which data and settings.
A machine learning pipeline is a defined sequence of steps that turns raw data into a deployable model: collecting and cleaning data, preparing features, training, evaluating, deploying and monitoring. Automating these steps makes results reproducible and lets new data be processed with little effort. A retailer, for instance, can retrain and test its sales forecasting model every week using the latest figures.
Kubeflow is an open-source platform for running machine learning workflows on Kubernetes, from data preparation through training and hyperparameter search to model serving. Originally initiated by Google, the project belongs to the Cloud Native Computing Foundation, where it reached “graduated” status in August 2026 (as of October 2026). It is aimed mainly at teams that train their own models at scale on their own or cloud infrastructure.
Amazon SageMaker is the Amazon Web Services (AWS) cloud service for building, training and deploying machine learning models. In late 2024 the original service was renamed “Amazon SageMaker AI”; since then, “Amazon SageMaker” has referred to a broader platform for data, analytics and AI with the SageMaker Unified Studio interface (as of October 2026). It is used mainly by data science teams running models on AWS infrastructure.
Model drift is the gradual decline in a model’s predictive quality during operation because reality changes compared with the training data. A distinction is made between changes in the input data (data drift) and changes in the relationship between inputs and outcomes (concept drift). Example: a fraud detection model misses new fraud schemes. Remedies include continuous monitoring of metrics and regular retraining.
