What is Agentic RAG?
LLM & AILevel: IntermediateAdvanced RAG Patterns
What is Agentic RAG?
Hints
- Break the query into candidate retrieval, context selection and generation.
- How would you locate the stage that lost a fact without using only one end-to-end metric?
Answer
Standard RAG is a fixed pipeline: query → retrieve → generate. Agentic RAG makes retrieval a dynamic tool:
User: "Summarize the key risks from Q4 reports of our top 3 competitors"
Agent:
Thought: I need to find Q4 reports for competitors A, B, C
Action: retrieve("Competitor A Q4 report")
Observation: [chunks]
Thought: Need B and C as well
Action: retrieve("Competitor B Q4 report")
...
Thought: Now I can synthesize
Answer: [synthesized summary with risks]
Agent decides whether to retrieve, what to search for, when to search again if results are insufficient. Key patterns: Self-RAG (model learns retrieval decision tokens), FLARE (retrieve when model is uncertain), ReAct with retrieval tool.
Source and licence: JustInCache & contributors — Awesome AI Interviews · Apache-2.0