Live demo · Machine learning

Forecasting with an honest backtest

A forecast with a prediction band, tested against simple baselines it has to beat.

Interactive demo on sample data. Runs in your browser. Not a client project. All live demos

Forecast horizon: 6 steps

actualmodel on held-out periodseasonal-naive baselineforecast with 95% band

What runs in this demo

Holt-Winters forecasting with parameters chosen by a small grid search, written for this demo and run in your browser.
A backtest on the last 12 points that compares the model with a naive and a seasonal-naive baseline.
A prediction band that widens with the horizon. You can also paste your own numbers.
The sample series are synthetic. If the model does not beat the baseline on your data, the demo says so.

What a partner project adds

Cleaning, missing-data handling and holiday or promotion effects built from the client's data.
Model selection across several methods, with rolling-origin backtests rather than a single split.
A scheduled pipeline, monitoring for drift, and an API or review screen for planners to override.

Built for: Software houses building planning, retail or operations products.

Try another demo

Ask a question, get the answer with the passage it came from, or a refusal.

Structured output, a confidence gate and an audit log around an AI step.

Paste a client brief and see which parts are AI work and what to clarify first.

Share a short brief. We’ll tell you whether the work is a fit and how a collaboration could be scoped.