Definition
A multi-stage retrieval strategy that indexes small, granular text segments linked to larger context blocks or document summaries, allowing the system to perform vector matches on precise snippets while feeding the LLM broader contextual surroundings. This architecture resolves the trade-off between high-precision search (small chunks) and semantic coherence (large blocks).
Not a folder-based taxonomy; it refers to the nested mapping of child embeddings to parent document nodes.
"A Microscope Zoom: The system uses a high-power lens to find a specific cell (chunk) but then switches to a low-power lens to view the whole organ (parent document) for context."
- Parent Document Retrieval(Component)
- Chunking(Prerequisite)
- Recursive Retrieval(Component)
- Context Window(Constraint)
Conceptual Overview
A multi-stage retrieval strategy that indexes small, granular text segments linked to larger context blocks or document summaries, allowing the system to perform vector matches on precise snippets while feeding the LLM broader contextual surroundings. This architecture resolves the trade-off between high-precision search (small chunks) and semantic coherence (large blocks).
Disambiguation
Not a folder-based taxonomy; it refers to the nested mapping of child embeddings to parent document nodes.
Visual Analog
A Microscope Zoom: The system uses a high-power lens to find a specific cell (chunk) but then switches to a low-power lens to view the whole organ (parent document) for context.