GenAI leverages the power of Large Language Models (LLMs), while protecting against their common pitfalls such as incorrectness & inappropriateness. This is done by using a Retrieval Augmented Generation framework (RAG).

The idea behind Retrieval Augmented Generation is that retrieval mechanisms and generative capabilities are used to enhance the performance of large language models. Practically this means that before an LLM is called to answer a query, first the relevant content is retrieved, such that it can be provided as source information to the LLM call.

Requirements

In order to be able to use the GenAI Engine properly, the following steps need to be taken:

• Knowledge needs to be provided by adding & synchronising resources (see these pages for more information on how to do this). This knowledge is chunked into smaller pieces of text and saved in an internal database.
• The required behaviour needs to be configured, e.g. by specifying the preferred language and tone of voice (see these pages for more information on how to set this up)

Approach

Now when a user asks a question to the chatbot, the GenAI Engine will perform the following steps:

• Retrieval: It will compare the incoming query with your organization's knowledge chunks, obtaining the three most relevant chunks and appending these together to create a single block of text with the most-relevant context.
• Rewriting: The context is provided to an LLM, which is instructed to formulate a response using the provided tone of voice, while using only the factual information present in the context.

---