A Language Model's ability to answer questions about specific documents depends heavily on how that information is prepared and presented. This chapter covers the initial stage of the data pipeline for Retrieval-Augmented Generation (RAG), focusing on transforming raw source material into a structured format suitable for processing.

You will learn how to:

• Load data from different sources, including PDF files and web pages, using the document module.
• Split a large document $D$ into a series of smaller text chunks $\{c_1, c_2, ..., c_n\}$ and apply different chunking strategies.
• Use the preprocessing module to clean and normalize text, which improves data quality for subsequent steps.

After completing these sections, you will have a clean, chunked dataset ready for the embedding process discussed in the next chapter.