AutoML Studio: train and explain ML models, now on GitHub

We're excited to introduce AutoML Studio, the newest addition to the splittingatom enterprise lineup — and the first one to launch fully open source.

AutoML Studio takes a prediction problem most teams already have — which customers will churn, which invoices will pay late, which leads will convert — and handles the machine learning end to end. Upload a table, pick the column you want to predict, and the studio selects the algorithms, trains the candidates, and evaluates them fairly on rows the model never saw.

Built to be understood

The difference is in how it reports back. Every metric comes with its meaning in plain English — how often predictions were right, how well the model tells categories apart, what to expect on new data. Results are benchmarked against simple baselines, so when a basic statistical model does nearly as well as the advanced one, AutoML Studio tells you that too.

From there you can:

• Explore what-if scenarios — adjust any input and watch the prediction update in real time, with a breakdown of which factors pushed it.
• Tune the decision cut-off — slide the confidence bar to trade off catching more cases against fewer false alarms, measured live on held-back test rows.
• Act on suggestions — retrain without columns that barely influence the model, then compare models to confirm nothing was lost.
• Take it with you — download the trained model and a shareable report.

Open source, available today

AutoML Studio is available now on GitHub — run it on your own infrastructure and keep your data where it belongs. Read more on the product page, or get in touch if you'd like help putting it to work.