Knowledge & Data · Software component
Per-Modality Fan-Out Retriever
Software componentKnowledge & DataKnowledge & Dataarc:PerModalityFanOutRetriever
A multimodal retriever that searches every modality-specific vector store in parallel, collecting each store's top-N candidates for cross-modal reranking.
Responsibility. Gathers candidate results in parallel from separate per-modality stores.
Also known as: Approach 3: Separate stores with cross-modal reranking
Variant of Multimodal Retriever abstract
When to choose. Choose for research or experimentation with best-in-class per-modality embedding models, or mature-MLOps production systems able to absorb the extra complexity and cost.
Relationships
invokes dependency
reads dependency
sends data to dynamic
alternative to variability
Design guidance
- SHOULD coordinate result ordering to stay consistent and fair across modalities with different retrieval characteristics.
Quantitative guidance
As stated by the sources; verify before use.
- Example: top 5 per modality across 3 modalities yields 15 candidates for reranking; M modalities x N candidates creates MN overhead (Ch2.7).
Classification
- Patterns
- Retrieve-then-rerankParallel fan-out retrieval
- Quality attributes
- Maintainability (ISO/IEC 25010)Functional suitability: correctness and validity (ISO/IEC 25010 | NIST AI RMF: valid)
Sources
- Ch2.7: T. Nguyen, "Multimodal RAG Approaches," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 2.7. ISBN: 9798244538229.