Abschlussarbeit

Beschreibung:

Conventionally organized customer support is effective, but labor-intensive and therefore costly. Chatbots based on Large Language Models (LLMs) have great potential to enhance companies’ customer service. Key challenges in their use include the quality of answers and concerns about data protection when processing sensitive information. This thesis aims to compare different implementation approaches of LLM chatbots that operate offline for data protection reasons. The support division of a company that sells software solutions to assist the development of mobile machinery will serve as a case study. Suitable LLMs are identified through literature review and then implemented. The software manuals, API descriptions and FAQs are used as a knowledge base. As part of a feasibility study, actual customer requests are anonymized and used as test cases to evaluate the response quality of the models. The answers are assessed by domain experts based on previously developed criteria. Subsequently, a statistical analysis is conducted and the thesis concludes with discussion of the results.