Cleanlab

Cleanlab automatically detects label errors, outliers, and other data quality problems in a training dataset, and ranks them by how much they're actually likely to be hurting your model rather than flagging everything equally. It's a data-centric approach to improving model reliability — fix the data instead of just throwing more model architecture at the problem.