HOW IT WORKS

From tables to predictions in five steps

No magic. Each step is inspectable, reproducible, and versioned.

The pipeline

1 · Load

[ TODO: describe data loading — CSV, Parquet, Postgres/MySQL sync, row-count preflight, preview. ]
2 · Clean

[ TODO: describe LLM-assisted cleaning, null handling, type inference, outlier detection. ]
3 · Detect schema

[ TODO: describe primary-key/foreign-key detection, time-column detection, review UI. ]
4 · Define task

[ TODO: describe natural-language task definition → SQL target, task-type inference. ]
5 · Train & evaluate

[ TODO: describe training loop, architecture selection, early stopping, checkpoint, metrics. ]

What's actually running under the hood — GNN, table encoders, task heads.

Deployment options, audit trails, compliance.

Sign up free. Train a real model in minutes.