Tag
A technique called the LLM Wiki pattern compiles raw documents into a structured wiki with an index, cutting Claude token usage by 95% by paying structural understanding costs only once during compilation instead of on every query.