What are the four main steps in a typical RAG pipeline?	Take a source document, break it into chunks, find relevant chunks for a query, and put them into a prompt.
Why is the process of chunking considered a complex and critical step in the RAG pipeline?	How the document is chunked has a huge impact on the quality and accuracy of the entire RAG pipeline.
Define Size-based chunking.	Dividing a large document into strings of text of equal, predetermined length.
What are the two primary downsides of simple size-based chunking?	Chunks often contain cutoff words and lack sufficient surrounding context (e.g., missing section headers).
How does the overlap strategy improve size-based chunking?	It includes a small amount of text from neighboring chunks, providing more context to each individual chunk.
Define Structure-based chunking.	Dividing text based on the overall organizational structure of the document, such as headers, paragraphs, or sections.
What is the main limitation of structure-based chunking?	It is difficult to implement reliably if the document does not have guaranteed, predictable formatting (e.g., plain PDFs).
Define Semantic-based chunking.	Using natural language processing (NLP) to group consecutive sentences or sections based on how related their meaning is.
What factors determine which chunking method should be used?	The specific use case and the guarantees available regarding the structure and content of the documents being ingested.
What is the risk of using an ineffective chunking strategy in RAG?	The pipeline may insert irrelevant or misleading context into the prompt, leading to significant errors in the final answer.
What are the default chunk settings mentioned in the transcript?	A chunk length of 150 with an overlap of 20.
What is the typical default output when using the "chunk by sentence" strategy?	Five sentences per chunk with one sentence of overlap.
When is the "chunk by section" strategy most effective?	When the document has a guaranteed and predictable structure (e.g., executive summary, section 1).
Why might the "chunk by section" strategy fail?	If there are no guarantees around the formatting of the document.
What is the final fallback chunking strategy when other methods fail?	Chunk by character.
Why is "chunk by sentence" often unsuitable for chunking code?	Code tends to use periods in unexpected places, leading to incorrect splits.
What is the primary factor that determines which chunking strategy to use?	The nature of the document and the guarantees available regarding its structure.
How can chunk quality be improved beyond the default settings?	By dramatically changing the chunk length and overlap values (e.g., length 500, overlap 150).