liveBook · Manning

Retrieve-then-generate pipeline

As explained in Designing AI Systems

Pipeline-based RAG is a retrieval-augmented generation approach in which retrieval runs as a fixed mechanical step for each request.

A retrieve-then-generate pipeline grounds each response by mechanically searching an organizational index with the raw user message, then placing the returned documents into the generation context. This fixed sequence works when the message is self-contained, but it cannot interpret conversational context, reformulate queries, decide whether retrieval is necessary, or retry with another search. section 10.5

Where the pipeline breaks

A request such as “What is our vacation policy?” can serve as its own search query. After discussing Q2 planning, “And what did the engineering team commit to?” lacks the keywords needed for reliable retrieval because the pipeline ignores conversation history. section 10.5

How Designing AI Systems covers it

How deep it goes
Implements indexing, search thresholds, document formatting, and context injection, but stops short of conversational query reformulation.
Examples
Uses vacation-policy lookup, Q2 planning outcomes, engineering commitments, and a company-knowledge index.
What it assumes
Assumes an indexed document collection with embeddings, chunking, metadata, and searchable organizational content.
What you can do after
Apply it via the Data Service to retrieve filtered documents and supply grounded context to generation.
How it connects
Contrasts fixed retrieval with agentic RAG, conversational context, query reformulation, and knowledge-base indexing.

Also explained in

Compare how 2 books explain retrieve-then-generate pipeline →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Designing AI Systems. The section links point to the parts of the book it draws on.