Towards a German-Language Chatbot for University Examination Regulations: A Comparative Study of Retrieval-Augmented Generation Systems

This repository contains the entire code of the bachelor thesis

The code is sorted by occurrence in the thesis as follows:

1. Small visualizations for embedding + code outputs
2. Some Web scraping
3. Dataset building in SQuAD-like format
4. PoC of Retrieval-Augmented Generation system with a custom pipeline
5. Evaluation with 4 Retrievers, 3 LLMs on 2 small Datasets
6. Evaluation results for Retriever and Reader in .csv files

The annotation_tool from haystack can be used to locally annotate txt data with labels and output it in SQuAD format