RAG Architectures Simplified

Ali Mirzaei

Nov 14, 2024

In this diagram, the traditional generation path for LLMs (similar to directly querying models like ChatGPT) is separated from the enhanced RAG (Retrieval Augmented Generation) pathway. This simplification shows how RAG systems integrate context-specific data (your data, which the model hasn’t seen during training) through targeted prompt engineering to enrich the prompt with the desired context.

Follow the blue lines to see how RAG systems dynamically incorporate external data, compared to the grey lines representing a basic, unaugmented generation process.

share/comment on