Agentic RAG: Retrieval as a Tool the Model Chooses

A linear, one-concept-per-lesson path through Agentic RAG, retrieval as a tool the model chooses to call rather than a fixed pipeline step, from Gemini's native function calling to a complete multi-tool FastAPI service built entirely from scratch. 26 lessons · 3 tiers.

Prerequisites: Comfortable writing basic Python. Assumes you've done this site's Naive RAG course (or otherwise know what dense retrieval and embeddings are) - this course is framework-free and specifically about retrieval-as-a-tool: when an agent should retrieve, how many times, and what else it should be able to call instead.

Lessons use Google's Gemini free tier (gemini-3.5-flash-lite for chat, no embedding model needed) via Gemini's native function calling (google-genai). No Docker, no database, no extra account - GOOGLE_API_KEY in a .env file is the only setup.

Built directly on google-genai, Gemini's native function-calling SDK, no agent framework in between.