Kayra is a free, open-source ML pipeline builder. You can use it as a Python library (pip install skyulf-core) or as a self-hosted web app with a visual canvas, training, tuning, a model registry and drift reports.

Before your pipeline trains, Kayra checks it for data leakage. If a step like an imputer, scaler or feature selector would learn from the test rows, it tells you which step and how to fix it.

When a model is ready, you can export it as a standalone project with its own API, or run your pipelines on Databricks with Kayra's bundle template.

Data scientists and ML engineers who use scikit-learn and want to keep their data on their own machine. It's still in active development (pre-1.0). For now, 0.9.7 released with many fix and improvements.

A plain scikit-learn Pipeline gives you the estimators. Kayra adds the leakage check, run history, comparison, deployment and drift reports around them. Unlike cloud ML platforms, it runs on your own hardware and nothing is uploaded. Unlike most drag-and-drop tools, every canvas pipeline can also run as plain Python.

Website: https://kayra.skyulf.com
GitHub: https://github.com/flyingriverhorse/Kayra
Demo: https://api.skyulf.com

Kayra

Open-source, self-hosted ML pipelines you can verify

Open-source, self-hosted ML pipelines you can verify