Question 1

What is Yung?

Accepted Answer

Yung is a data sanitation layer that sits between labelled data and model training. It identifies mislabelled samples and data points that actively harm model performance, removing them before they ever reach production.

Question 2

How does influence scoring work?

Accepted Answer

Yung uses influence-function mathematics to compute how much each individual data point helps or hurts the model's validation loss. Samples with high harmful influence are flagged and removed before retraining.

Question 3

What models does Yung support?

Accepted Answer

Yung is model-architecture-agnostic. It works with LightGBM, neural nets, XGBoost, and any other downstream model at materially lower compute cost.

Question 4

How is Yung delivered?

Accepted Answer

Yung is delivered as a trial API that customers can test independently before signing engagements, with a self-service platform. The core requires only PyTorch and NumPy.

Shift from volume-first to a precision-first strategy.