Raw text data, particularly from large sources like web crawls, requires significant refinement before it can be effectively used for training large language models. This chapter addresses the practical steps involved in cleaning and structuring these datasets. We will cover methods for filtering out low-quality content, normalizing text representations, removing extraneous material such as HTML tags and navigation elements, identifying and handling duplicate documents, and isolating text in target languages. Finally, we'll look at structuring these operations into scalable data processing pipelines suitable for handling very large volumes of data.