Why RAG is Important: The Knowledge Bridge
Understand the 'Knowledge Cutoff' problem. Learn how RAG (Retrieval Augmented Generation) connects Gemini to your private, real-time data.
Why RAG is Important
Gemini knows a lot, but it doesn't know your secrets.
- Problem: It doesn't know your company's Q3 sales data (private). It doesn't know the news from 5 minutes ago (cutoff).
- Solution: RAG.
The Concept
RAG is an open-book test.
- Retrieval: You act as the librarian. You find the exact page in your internal wiki that answers the question.
- Augmentation: You paste that page into the Prompt.
- Generation: You ask Gemini "Based on the text above, answer the question."
RAG vs Context Window
We learned Gemini has a 1M token window. Why use RAG?
- Cost: Filling 1M tokens costs money ($$) per query. RAG sends only 2k tokens ($).
- Latency: 1M tokens takes 30s-60s. RAG takes 2s.
- Scale: If you have 100 Million documents (Terabytes), even Gemini's context window is too small.
Summary
Use RAG for Large Databases and Low, Latency. Use Long Context for Single Huge Documents.
In the next lesson, we build the engine of RAG: Vector Stores.