A RAG system is only as good as the text it retrieves, and that text starts as messy real-world files: PDFs, web pages, slides, and notes. Before anything can be embedded, documents have to be loaded, cleaned, and split into pieces small enough to search and to fit in a prompt. How you split them has a big effect on whether the right passage is found later.
The lessons start with an overview of chunking: what chunks are, and how chunk size and overlap affect retrieval. You will then see how to load documents and split text in code, compare common chunking strategies, and look at parsing complex documents such as PDFs with LlamaParse. A final long lesson walks through five increasingly sophisticated ways to split text.
Watch the lessons in order. The last video is long and goes well beyond the basics, so feel free to watch just the beginning for now and come back to it later.
An overview of why documents are split into chunks, the main chunking strategies, and how choices like chunk size and overlap change what gets retrieved.
Video 14minA hands-on walkthrough of loading documents and splitting them into chunks with LangChain's document loaders and text splitters.
Video 3minMervin Praison compares different ways of chunking data for RAG and how the choice affects the quality of the answers a model gives.
Video 14minHow to parse complex documents such as PDFs into clean, structured text with LlamaParse, so that the content going into your RAG pipeline is usable.
Video 1hGreg Kamradt's long walkthrough of five approaches to splitting text, from simple to advanced. The first levels give useful background, and the rest is an optional deep dive.
Video 1h 9min