An LLM-powered AI assistant designed to support international students seeking information about studying in Finland
Year: 2026
Type: dissertation
Abstract: This study examined the design, implementation, and evaluation of a locally deployed retrieval-augmented generation (RAG) assistant for international students seeking information about studying and living in Finland. Institutional information relevant to students is often fragmented across multiple official sources, making it harder to find reliable answers to administrative and practical questions. The objective of the thesis was to investigate whether a locally deployed assistant could provide grounded and reliable responses within a clearly defined information domain. A prototype system was developed using a Streamlit frontend, a FastAPI backend, a Milvus vector database, and a locally served language model through Ollama. The system was evaluated through controlled testing focused on factual grounding, retrieval effectiveness, local runtime performance, and a small feedback survey related to reliability and usability. The study showed that retrieval improved factual control compared with relying on generation alone, although it did not fully eliminate unsupported output. Model choice and retrieval configuration both affected system performance, and Qwen 2.5 7B provided the most balanced results under the selected local setup. The results indicate that a locally deployed RAG-based assistant can provide reliable domain-specific support when grounded in trusted sources, although its effectiveness depends on retrieval quality, model behavior, and maintenance of current source material.
Source: Theseus (Ammattikorkeakoulujen)
Author: Bahgat, Mohamed
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: green
OpenAlex ID: W7160005773