seldon-core

Why we included this project

Platform teams running machine learning models in production on Kubernetes will find Seldon Core a mature way to standardize that work. It packages trained models as inference services you can deploy, monitor, and scale through a single operator, and it supports a wide range of model types rather than locking you into one runtime. On the serving side, it can consolidate multiple models on shared infrastructure and autoscale them, which keeps infrastructure costs down. Its experiment features route traffic between candidate models for A/B tests or shadow deployments. And the Kafka-based pipelines go further, chaining models and custom components into a realtime application instead of serving each model in isolation. If your platform team wants one reproducible path from trained models into production, this is a solid place to start.