RAG Chunking Strategies: Recursive Text vs Semantic vs Markdown Header

90% of RAG retrieval failures originate from poor document chunking. Match your chunker to your document anatomy:

1. Technical Markdown & Documentation Use `MarkdownHeaderTextSplitter` so each section retains its structural hierarchy:

from langchain_text_splitters import MarkdownHeaderTextSplitter
headers_to_split_on = [("#", "H1"), ("##", "H2"), ("###", "H3")]
splitter = MarkdownHeaderTextSplitter(headers_to_split_on=headers_to_split_on)

2. Dense Analytical Reports Use `RecursiveCharacterTextSplitter` with `chunk_size=600` and `chunk_overlap=120` across separators `["\n\n", "\n", ". ", " "]`.

The Golden Rule Never separate code snippets or data tables across chunk boundaries. Always attach parent section metadata to downstream vector embeddings.