Anwendungsbeispiele von Retrieval Augmented Generation bei Miba AG

Abstract

When using language models, users are faced with the challenge that language models suffer from hallucinations on the one hand and do not have up-to-date information on the other. Retrieval Augmented Generation is a new approach to reduce the problem of hallucinations in language models and to provide relevant and up-to-date information for solving tasks. This thesis deals specifically with the collection of use cases for Retrieval Augmented Generation from the environment of the supply chain department at Miba AG and the elaboration of potential challenges. In preparation for this, a definition of general terms in the context of artificial intelligence is done based on specialist literature as a foundation for the following chapters. A review of past progresses is also performed to draw conclusions for the future development. The basics of language models, how they work, the challenges and possible solutions are elaborated. It is important to understand the general concept, as retrieval augmented generation is an extension of it. Retrieval augmented generation is a technique in which the original prompt is expanded with information from a knowledge database and passed on to the language model. The language model thus not only has the user input, but also contextual information. It therefore not only has the user input, but also context-relevant information. Application examples show the potential. In the basics of implementation, the most important concepts for creating a project brief and the necessity of data governance are discussed. The expert interview was chosen as the method for collecting the use cases. A total of 13 use cases were recorded, evaluated, and selected for further processing. The result is a defined project assignment based on the information from the specialist literature and the findings from the interviews, considering the requirements and challenges that were jointly identified.

Date of Award	2024
Original language	German (Austria)
Supervisor	Klaus Arthofer (Supervisor)