Generative AI (GenAI) is in the house, and it is here to stay. You might ask, so how can my business make use of it? A great example is an Enterprise Chatbot, where any user can ask questions of their business, and get the appropriate response. However pre-trained Large Language Models (LLM) alone are not enough, it needs your companies data! Which is where RAG steps in.
Delivering a RAG system with Zetaris
What is RAG?
RAG stands for Retrieval Augmented Generation, which is an approach used in large language models to enhance their ability by accessing and utilizing external information when answering questions or performing tasks.
The "Retrieval" part involves retrieving relevant information from external sources like websites, databases, datalakes, knowledge bases etc. This allows the model to go beyond its pre-trained knowledge.
The "Augmented" part takes the retrieved information and conditions or primes the language model with this extra context before generating an output.
The "Generation" part is where the language model uses its natural language abilities along with the augmented context to produce a final output like an answer, analysis, or generated text.
Why is it important?
RAG is useful because while large language models have a lot of general knowledge from pre-training, they can't know everything about your business and having mechanisms to quickly retrieve and leverage relevant external data allows them to produce more up-to-date, complete and accurate outputs, especially for queries requiring very specific or latest information not contained in the training data.
In the business context, your enterprises varied landscape of datasources and systems is what your RAG should be built from, with it, it will allow you to deploy Natural Language applications, like ChatBots, Virtual Assistants and Natural Language Analytic agents.
When done properly, users from all industries will see benefit, from health, to automotive, to finance and retail. Being able to query your enterprise's full set of data allows a nurse or a clinician, to ask questions like,
"How many beds are available in Ward F right now, and where are they?"
and get a generated response to say,
"There are 4 beds currently available, they are 113, 129, 189 and 190".
Or for an analyst or manager, could quickly ask,
"What is the average productivity percentage of my team for the week, and how does that compare to our target? Are there any outliers?
and get a generated response to say,
"The teams average is at 78%, which is lower than our 85% target. Grace is at 95% for the week, and George is only on 60%."
Delivering a RAG system with Zetaris
Being able to deliver a RAG system to an enterprise is not easily done without the right tools. The best RAG, will consider all data sources that are related to the question being asked, this is where Zetaris steps in.
Zetaris uses a decentralized approach to access, prepare and deliver your data, at scale. Fundamentally, Zetaris can connect to a range of data sources, of varying types, so that your organisation has a single access point, or source of truth, for their data.
Take a customer for example, the insight of a customer might span across your companies warehouse, along with a CRM platform, a legacy system, and even a combination of them. Zetaris can connect to each one of those platform, and build consistent reference model that your organisation can refer to, see Customer_360 Semantic Layer below,
Customer_360 Semantic Layer
Then in the context of RAG, this Virtual Semantic Layer is what we will build our RAG from to give your LLM the most contextual reference point it could need.
Zetaris comes with its own CoPilot, called GenZ, that you can plug into your deployment with varying levels of customisation around which LLM to use and where it is located, either public or private. The high-level design is shown below.
RAG with Zetaris
Reach out to us today for install options.