A RAG Primer: Using AI with Your Own Content Part 1

Drawback of using AI model chat version (plain vanilla version)

In our past articles, We’ve explored how to build a SmartGPT for use within your Enterprise. EnterpriseChatAssistant or EnterpriseCoPilot Chatbot offers a great tool for your internal marketing or sales teams to prompt and use it for personal efficiency like an ideation buddy or a productivity buddy. This is a great way to tap into the collective knowledge an AI model has to get better at your work and improve productivity.

However, in an enterprise or a business we want to use not only the specific knowledge that the AI LLM model has from its training data, but also knowledge of your specific environment and data.

For example, let's say you work for GE. The LLM models (OpenAI, Claude, Gemini, Llama etc) may know about X-ray scanners but you want it to know specifically about GE’s X-ray scanners in greater detail. This allows you to complement the LLM model capabilities with additional information specific to your domain and your company’s products so it learns more about the specific context you are operating in.

Or perhaps you are a B2B Marketing manager or a Sales Operations manager and you know how to use an AI model to create a generic email campaign (using either Microsoft Copilot, ChatGPT, Claude, or Gemini). However, it would be incredibly useful if the LLM model knew the specifics of your products, the persona you have targeted in the past and perhaps had a look into your product’s positioning. This information is typically available within documents within your Sharepoint, Google Drive, or some shared repository behind your company’s firewall. But it is not available to LLMs in their native versions.

Hence, the general problem while using LLMs is that while we are asking it to use its existing knowledge, we also want it to know or “learn” about all the unstructured and enterprise data available.

A few questions typically come by when trying to use AI models in enterprise scenarios:

- How do we get it to learn or be aware of all that and make it hyper-useful in the enterprise context?(an example could be your specific for your sales and marketing campaign)

- What if we could equip it with additional data that is specific to our business and make it aware of our domain? Then we could make the AI model create content that would “learn” from our documentation and be hyper-useful.

The smart Marketing and Sales Operations team members want the capabilities of the AI but they want it to be used on their own data and context/domain.

- They want the AI Model to learn their enterprise data and content and then make use of it to help them in their job.

- They want it to be used on the structured and unstructured data available within the organization.

- They know that LLMs can read their Sharepoint & Google Drive content and surface the unique insights. Thus leveraging that existing data to make their sales and marketing activities more impactful.

Over the past year or so, smart people working on this domain have found that they can solve this by using a technique called as RAG or Retrieval Augmented Generation. RAG is a technique that combines or connects generative AI services to external resources.

While this is a lot of acronyms for the typical sales and marketing professional, think of this as the underlying technique that allows you to utilize your own data with the smarts of an AI model.

What this technique implies is that we augment the LLM model’s inherent knowledge with additional data available within our company. That allows the model to “read”, “see” and absorb the local files, brochures, training materials, manuals, and knowledge bases to frame up a response that is grounded in the company’s data.

In layman’s terms: you are aiming to use the capabilities and existing knowledge of the LLMs, “train” it with your data and then utilize it in your company so it can give you your company specific answers or perform tasks.

In the next part of this article, we will look at RAG in detail.