# Retrieval-Augmented Generation

> A design pattern that retrieves relevant documents for a question and passes them to the model, so the answer can cite its sources.

Record: `architecture:retrieval-augmented-generation`
Kind: pattern
Layer: [Correctness Core](https://banes-lab.com/records/layer/correctness-core.md)
Severity: contextual
Scope: language-model system, knowledge retrieval
Aliases: RAG
Canonical: https://banes-lab.com/ontology#architecture-retrieval-augmented-generation

Listed in [Architecture principles](https://banes-lab.com/api/records/architecture.md), after [Model Inference](https://banes-lab.com/records/architecture/model-inference.md) and before [Vector Search](https://banes-lab.com/records/architecture/vector-search.md).

## Repair

- Refactored by: Add Retriever, Add Reranker, Add Citation Grounding
- Detected by: missing citations/context in grounded tasks
- Violated by: answers generated without relevant retrieved context where required
- Measured by: retrieval precision/recall, groundedness
- Enforced by: RAG evals

## Requires

- [Retriever](https://banes-lab.com/records/lexicon/retriever.md)
- [Document Store](https://banes-lab.com/records/lexicon/document-store.md)
- [Grounding Strategy](https://banes-lab.com/records/lexicon/grounding-strategy.md)

## Reinforces

- [Explainability](https://banes-lab.com/records/architecture/explainability.md)
- [Knowledge Freshness](https://banes-lab.com/records/lexicon/knowledge-freshness.md)

## Enables

- [Contextual Generation](https://banes-lab.com/records/lexicon/contextual-generation.md)

## Conflicts with

- [Ungrounded Generation](https://banes-lab.com/records/lexicon/ungrounded-generation.md)

## In tension with

- [Retrieval Quality/Latency](https://banes-lab.com/records/lexicon/retrieval-quality-latency.md)

## Tensions

- [Retrieval-Augmented Generation (RAG) / Retrieval Quality/Latency](https://banes-lab.com/records/tension/retrieval-augmented-generation-rag-retrieval-quality-latency.md)

## Severity

- [contextual](https://banes-lab.com/records/vocabulary/severity-contextual.md)

## Category

- [Model Architecture](https://banes-lab.com/records/architecture-category/model-architecture.md)

## Reinforced by

- [Vector Search](https://banes-lab.com/records/architecture/vector-search.md)

## Linked from

- [Systems built around a model](https://banes-lab.com/software-architecture/scale/systems-built-around-a-model.md)
- [Model Architecture](https://banes-lab.com/ontology/principles/architecture-category-model-architecture.md)
- [Model Architecture](https://banes-lab.com/ontology/lexicon/lexicon-category-model-architecture.md)
- [Severity levels](https://banes-lab.com/ontology/schema/the-vocabulary-severity.md)
- [The resolutions](https://banes-lab.com/ontology/schema/the-resolutions.md)
