01
RAG Pipeline
A retrieval-augmented system first identifies potentially relevant external information and then conditions generation on that information.
q → retrieve(D,q) → context(C) → P(y|q,C)
02
Engineering Variables
- Chunk size and overlap.
- Embedding model.
- Indexing strategy.
- Retriever recall.
- Reranking.
- Context-window allocation.
- Citation and provenance.
- Freshness and document lifecycle.