RAG KI – Enrich language models with company data

RAG – WHAT IS THIS?

RAG is a system that forwards an incoming question to the LLM. First, however, a specially prepared vector database is searched for all data relevant to this question. These are then transmitted to the LLM by means of prompts.

The LLM therefore not only has to rely exclusively on its trained knowledge, but also receives all the relevant data with the question.

HOW? – TWO PROCESSES MAKE IT POSSIBLE

The data is prepared at the beginning. The data is stored in a vector database and, in most cases, normalized and split.
In most cases, it is also processed, normalized and split. In addition, data such as writing texts, program code, images or audio require their own processing.

In the second process, the user’s question is answered. Here, the RAG accesses the vector database and searches for the relevant data that can be found thanks to the previous preparation of the data. The data is then transformed once again to provide a meaningful context for the LLM.

WHAT DOES THE VECTOR DATABASE HAVE TO DO WITH IT?

The special feature of the RAG system is the vectorization of the data and the storage in a vector database. This contrasts with “normal” rational databases, which are stored in tabular form. While the queries here are based on the values of the entries themselves, a vector database is queried according to the number of data entries that resemble a data entry. This allows the RAG system to find the text passages in its vector database that are most relevant to the question. These relevant text passages in the documents are then forwarded by the RAG to the LLM.

PROMPT ENGINEERING

Prompt engineering refers to the input that is given to an LLM. In relation to a RAG system, this refers to the user’s question, the context and the instructions that are passed to the LLM to generate the response. A corresponding prompt is required so that each step of the implementation in the RAG is taken over by an LLM.

RAG FEATURED BY MEDIENWERFT

In order to always be able to offer our customers state-of-the-art solutions in e-commerce, the topic of AI with RAG was on our agenda early on. Even before the first customer inquiries, we started developing a RAG system for internal use. This enables us to use the extensive knowledge that we have accumulated over the years in the form of tickets. In essence, this enables structured access to our internal know-how.

Would you like to find out what benefits your company can gain from having its own RAG system?

Then let us advise you now!