Xeplr Pro Suite

Xeplr AutoML

Pick what you want to predict, point each input at a column you already have, and train. The prediction lands in a real table that registers itself as an ordinary Dataset — so you chart it the same way you chart anything else.

Prediction as a six-month detour

The distance between "we should forecast this" and a model in production is usually a hiring plan. Most of that distance is not modelling — it is working out which inputs are needed and whether the data even exists.

How it works

Objectives carry their own requirements

An objective declares the features it needs and how badly it needs them, so the wizard can tell you what is missing before anything is trained rather than after.

No algorithm to choose

Nothing here invents a model. An objective names an established Python library — Prophet, for forecasting — and the product’s job is to make picking and configuring it invisible.

Validated before a row is read

Feature maps are checked structurally first, so a run fails on a missing input immediately instead of after pulling data and starting Python.

The output is just a table

Predictions are written into a real target table which registers as a Dataset. There is no separate AutoML output screen, on purpose — history and forecast sit in one table so a single chart shows actual vs predicted.

The wrong column is obvious before training

As each input is pointed at a column, the screen shows a live sample of its real values, and training then runs in the background with a live log.

A crash never reads as success

Python reports back one line at a time, and only an explicit success or error line decides the outcome. A process that exits cleanly without saying it succeeded has not succeeded.

What it does not do yet

Current as of the product readme. If something here matters to you, ask — some of it is next.

  • Every mapped feature must come from the same table — there is no cross-source join here.
  • Sourcing is from an existing table only. Formula, upload and API sourcing are not built.
  • No scheduled retraining, and no accuracy shown in the UI yet. The accuracy figure recorded is in-sample, not a held-out backtest.
  • No input that needs a future value, such as planned promotion spend — there is no screen to collect projected values.

See it against your own data

A demo runs on your schema, not a sample dataset. Thirty minutes, no deck.