Enhancing an AI Chatbot for Student Support Services Through Official University Knowledge and English–Chichewa Multilingual Support

Student support offices in higher education institutions handle recurring questions about admissions, fees, registration, examinations, library services, and other university processes. When institutional information is distributed across channels or staff are unavailable, students may experience delays in finding reliable answers. This paper describes the design, implementation, and preliminary evaluation of UniBot, an AI-powered student support chatbot developed for DMI St. John the Baptist University. The system combines a lexical FAQ-matching stage with a generative response stage that is constrained by active, approved university FAQ information. It records source metadata and displays supporting FAQ references for eligible responses. The interface also provides English and Chichewa language modes and browser-based voice-to-text input.
The work was evaluated through automated software tests, browser-based checks, synthetic FAQ-matcher query sets, and observations of selected deployed interactions. The preliminary matcher results reported for a 115-query synthetic development set show an increase in precision from 63.0% to 92.2%, recall from 63.0% to 72.8%, accuracy from 60.0% to 76.5%, and a reduction in false-positive rate from 47.1% to 14.7%. These figures are preliminary: a 38-query holdout set was subsequently used during tuning and is therefore not an untouched independent test. The implementation test suite was reported to pass 592 of 592 tests, with additional browser checks reported separately.
The results demonstrate implemented software capabilities and promising preliminary FAQ-matching behavior, but do not establish answer correctness across real student questions, Chichewa translation quality, student satisfaction, or institutional impact. The paper documents the architecture, current behavior, limitations, and further evaluation required before broader deployment.
Keywords: student support chatbot; artificial intelligence; FAQ matching; grounded question answering; source attribution; Chichewa; higher education; voice-to-text.