Definition
The process of programmatically rewriting, expanding, or decomposing a user's raw input into an optimized format to improve retrieval precision and agent reasoning. While it significantly boosts retrieval quality (Recall/NDCG), it introduces trade-offs in the form of increased inference latency and higher token costs due to the additional LLM call required before retrieval.
Distinct from simple spell-check; it focuses on semantic alignment with the underlying vector database or tool API schemas.
"A blurry photograph being processed through a sharpening filter to reveal hidden details for an image recognition system."
- HyDE (Hypothetical Document Embeddings)(Strategy)
- Query Expansion(Component)
- Semantic Search(Prerequisite)
- Chain-of-Thought(Component)
Conceptual Overview
The process of programmatically rewriting, expanding, or decomposing a user's raw input into an optimized format to improve retrieval precision and agent reasoning. While it significantly boosts retrieval quality (Recall/NDCG), it introduces trade-offs in the form of increased inference latency and higher token costs due to the additional LLM call required before retrieval.
Disambiguation
Distinct from simple spell-check; it focuses on semantic alignment with the underlying vector database or tool API schemas.
Visual Analog
A blurry photograph being processed through a sharpening filter to reveal hidden details for an image recognition system.