Geek-Out Time: RAG vs. CAG in Google Colab

We often talk about enhancing language models with external knowledge. Two common strategies:

In this geek-out session, I played with two popular strategies for enhancing language model generation: Retrieval-Augmented Generation (RAG) and Context-Aware Generation (CAG). Using Google Colab, flan-t5-base , and a small FAISS-backed document store, I ran a controlled benchmark to compare their behavior, output quality, and performance. The goal? See how each approach handles the same question under identical conditions — and whether theory holds up in practice.

The Setup

The Full Code (Google Colab-Ready)

This is an excerpt — the full article continues on Medium.