Knowledge & Data · Software component
Hybrid Retriever
Software componentKnowledge & DataKnowledge & DataVariation point (abstract)arc:HybridRetriever
A retriever that combines vector similarity search with knowledge-graph traversal, linking entities found in retrieved text to graph nodes.
Responsibility. Combines vector search with graph traversal for queries needing both.
Also known as: Hybrid RAG+KG, GraphRAG, Hybrid RAG retriever
Variant of Retriever abstract
When to choose. Choose when queries need both semantic similarity and relationship traversal.
Variants
| Variant | When to choose |
|---|---|
| Graph-Constrained Vector Retriever | — |
| Graph-Enhanced Retriever | Choose when retrieved chunks need relationship context invisible to vector search (authorship, employment history, contradictions between documents). |
| Parallel Fusion Retriever | Choose when you cannot predict whether vector search or the graph will surface the needed information, trading complexity for completeness. |
| Retrieval-Augmented Graph Retriever | Choose when the graph is too large for exhaustive traversal (millions of nodes, billions of edges) and vector search can identify the entities around which to scope traversal. |
Relationships
invokes dependency
- Entity Linker abstract Ch1.7A
- Graph Retriever Ch1.7A
- Vector Retriever abstract Ch1.7A
is cached by dependency
is invoked by dependency
- Agent Controller abstract Ch1.8
reads dependency
- Knowledge Graph Store abstract Ch1.8
- Vector Index Store abstract Ch1.8
alternative to variability
- Graph Retriever Ch1.7A Ch5.7 +1
- Vector Retriever abstract Ch1.7A Ch5.7 +1
Design guidance
- SHOULD be adopted only when queries consistently need both semantic understanding and precise relationship tracking (e.g., financial compliance, fraud detection, research analysis).
- SHOULD run vector search and graph queries asynchronously in parallel where possible to limit accumulated latency.
- MUST keep entity references synchronized between the vector store and the graph.
- SHOULD be adopted when semantic and structural signals both matter; deployments often start with pure vector stores and migrate to hybrid as retrieval sophistication requirements grow.
- SHOULD be justified by requirements, since it duplicates engineering for embeddings and graph extraction and requires cross-store consistency.
Quantitative guidance
As stated by the sources; verify before use.
- p95 latency ~180ms vs ~120ms vector-only (+50%) due to sequential vector search, entity extraction, graph expansion, and synthesis (Ch1.7B).
Classification
- Patterns
- Vector search then entity linking then graph expansionHybrid RAG
- Quality attributes
- Functional suitability: correctness and validity (ISO/IEC 25010 | NIST AI RMF: valid)
- Risks mitigated
- Missed multi-hop connections (pure RAG)Missed fuzzy concepts such as 'similar markets' (pure graph)
Sources
- Ch1.7A: T. Nguyen, "Relational Reasoning with Knowledge Graphs - The Fundamentals, Integration, and Extraction," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 1.7A. ISBN: 9798244538229.
- Ch1.7B: T. Nguyen, "Relational Reasoning with Knowledge Graphs - Hybrid RAG+KG Integration," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 1.7B. ISBN: 9798244538229.
- Ch1.8: T. Nguyen, "Scalability and Production Deployment," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 1.8. ISBN: 9798244538229.
- Ch5.7: T. Nguyen, "Episodic Memory," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 5.7. ISBN: 9798244538229.
- Ch5.8: T. Nguyen, "Semantic Memory," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 5.8. ISBN: 9798244538229.