Course Wiki

Agentic RAG

Letting the model control its own retrieval: deciding when to search, what to search for, and whether the answer is good enough.

Agentic RAG hands the retrieval step to the model. Instead of a fixed pipeline that always searches once and answers, the LLM is given search as a tool it can call: it reformulates the question, decides whether results are relevant, searches again with different terms, and answers when it has enough context — all through function calling.

The course frames this as the bridge from a fixed RAG pipeline to an agent: the retrieval logic moves from your code into the model’s loop. The same search tools, prompts, and knowledge base serve the agentic version, which is why the module comes first.

Taught in

• LLM Zoomcamp

  • Module 1: Agentic RAG
    • Introduction
    • Agents
    • Function Calling
    • The Agentic Loop
  • Module 3: Orchestration
    • AI Agents

• AI Dev Tools Zoomcamp

  • Module 5: Coding Agent Building Blocks: Reusable Skills and Specialized Subagents
    • Module 5 — Coding Agent Capabilities: MCP, Skills, Plugins, and Custom Agents ## Related concepts

• RAG

• Function Calling

• Vector Search

• LLM Evaluation