AI IntelligenceAI-AGENT.NEWS

AI Engineering Topics

Retrieval-Augmented Generation (RAG)

RAG grounds a model response in material retrieved for the current question instead of relying only on model training.

170 matching public evidence items

All Topics · Agent Systems · Retrieval and Knowledge

How it works

A query finds relevant chunks, ranking selects the strongest evidence, and the agent answers from that bounded context with source links where possible.

  1. Question
  2. Retrieve
  3. Rerank
  4. Answer with evidence

Key concepts

Chunking
Dividing source material into retrievable units.
Retriever
The component that finds candidate evidence.
Reranker
A second-stage scorer that improves evidence ordering.

Engineering risks

  • Poor chunks can separate claims from necessary context.
  • Large evidence bundles increase cost and dilute relevance.
  • Answer quality must be evaluated separately from retrieval quality.

Aggregated from the site

Evidence and developments

Matching public material uses the same topic rules as AI Engineering.