Chunking
Splitting documents into smaller passages before embedding them for RAG. Chunk size trades retrieval precision against context completeness, and directly affects vector count and cost.
What it is
Each chunk becomes one vector in the vector database. Small chunks (100-200 tokens) give sharp matches but may miss surrounding context; large chunks (500-1,000) carry more context but dilute relevance and cost more to send to the LLM.
Why it matters
Halving chunk size roughly doubles vector count and embedding cost, and changes how many chunks you must retrieve to cover an answer. It is a tuning knob with a direct line to the bill.