From your data to a specialist's training set.

Paste a labelled export — support tickets with a category, deals with an outcome. fallkit finds the label, turns each row into an input → output pair, and splits them into train and a held-out eval so nothing leaks — because a gate verdict means nothing if the model was tested on what it trained on. If there's no label, it says so and stops. It never invents one. Everything runs locally; the pairs stay on your machine, ready for fallforgemint.