Contextual Compression

Retrieved chunks are 90% noise, 10% gold. Before sending to the LLM, you extract only the relevant parts. What happens to the chunks? Flag format: CONGRESS{[what_happens]} Example: CONGRESS{translate_to_english}