Understanding Retrieval Augmented Generation (RAG): Is This New Era for Prompt Engineering?

Understanding Retrieval Augmented Generation (RAG): Is This New Era for Prompt Engineering?

Bridging the Gap Between Information Retrieval and AI Creativity

Marko Vidrih
4 min readNov 3, 2023


In the rapidly evolving domain of artificial intelligence (AI), keeping abreast with the latest advancements is crucial for both practitioners and enthusiasts. A term that’s recently been stirring conversations is “Retrieval Augmented Generation,” abbreviated as RAG. This technology represents a watershed in the way large language models (LLMs) process and generate information. Below I delve into what RAG is, its significance, and the implications it carries for the future of AI-driven communication.

What is Retrieval Augmented Generation (RAG)?

RAG is a hybrid AI approach that merges two traditionally separate aspects of machine learning: retrieval-based and generative models. In essence, it’s about equipping a base LLM with an external database from which it can dynamically fetch information to inform its outputs. This doesn’t just expand the model’s knowledge horizon; it fundamentally changes how it interacts with queries.

How Does RAG Work?

A typical RAG system works in two stages:

  1. Retrieval Stage: When prompted with a query, the RAG system first scours its connected database to find relevant information. This database could range from a curated vector database to the expansive realms of the internet.
  2. Generation Stage: Once the relevant data is retrieved, the generative component of the model synthesizes this information, integrating it with what it has learned during its training to create coherent, contextually rich responses.

Reducing Hallucinations in AI

One of the touted benefits of RAG is its propensity to reduce “hallucinations” — instances where AI models generate false or irrelevant information. By relying on up-to-date and context-specific data retrieved during the process, RAG models anchor their responses in reality more firmly than generative models that rely solely on their training data.

Is RAG Just Advanced Prompt Engineering?

The conversation around RAG has sometimes boiled down to it being an advanced form of prompt engineering. Indeed, at its core, RAG optimizes how context is presented to a model to elicit the best possible response. However, this belies the architectural and functional complexity that RAG introduces, transcending the traditional limitations of prompt engineering by incorporating live retrieval processes.

Prompt Engineering

The Comparative Edge of RAG

The AI community has begun to notice that RAG can match, and at times surpass, the performance of models with significantly longer context windows. This observation was supported by studies showing that a RAG system, when paired with a 32k context window LLM, can outperform similar models with full context availability. This suggests that RAG’s method of contextualizing responses can be more effective than simply providing more pre-existing context.

Introducing Self-RAG

A novel evolution of RAG is Self-RAG, an approach aimed at teaching LLMs to discern when and how to retrieve information more accurately. This involves the generation of a specialized critique dataset that informs the model of optimal retrieval times and sources, improving the overall relevance and accuracy of the generated content.

Building RAG Systems

Developing a RAG system requires a nuanced understanding of AI engineering, specifically the ability to craft prompts that guide the AI to optimal information retrieval and response generation. Prompt engineers must be skilled in creating and managing the intricate web of prompts that constitute a RAG’s operational framework.

Challenges and Future Prospects

While RAG significantly enhances the capabilities of LLMs, it isn’t without its challenges. The potential for retrieving bad responses or misinformation remains a concern that necessitates the ongoing refinement of RAG systems. Moreover, as AI continues to evolve, learning how to develop and utilize RAG effectively will be an essential skill for AI engineers.


Retrieval Augmented Generation is revolutionizing the landscape of AI communication, offering a new frontier where information retrieval and generative creativity intersect. By understanding and harnessing RAG, AI practitioners can create models that not only converse with unprecedented accuracy and relevance but also pave the way for more sophisticated AI applications in the future. As we stand on the cusp of these developments, it’s clear that RAG systems will play a pivotal role in shaping the next generation of AI-powered interactions.

Follow me on social media



Project I’m currently working on




Marko Vidrih

Most writers waste tremendous words to say nothing. I’m not one of them.