Design and Evaluation of an Open-Source, Locally Deployed Chatbot System for Higher Education

Large language models (LLMs) integrated into chatbots offer new opportunities for personalized learning and student support in higher education. However, reliance on commercial LLM-based systems raises concerns around privacy, transparency, and equitable access. We present an open-source, locally deployed chatbot system designed to align with institutional values of autonomy, fairness, and academic integrity. Based on the open-source tools Open WebUI and Ollama, the system uses a model-agnostic architecture and provides features such as document upload and analysis, web search, and LDAP-based authentication. Manual evaluation by two human evaluators with overlapping expertise across seven key functional areas using 15 test cases showed a 73.3 % compliance rate, with strong performance in summarization, ethical safeguards, and prompt filtering. Limitations were found in web search reliability and gender-neutral text generation. The evaluation framework is extensible and may be used to benchmark both open-source and commercial chatbot systems. Planned future work includes formal evaluation of user adoption and usability in real educational settings.