Taming Technical Bias in Machine Learning Pipelines