Year: 2025
Type: dissertation
Abstract: This thesis developed an AI advisor system for the Arctic Valley business simulation at Lapland University of Applied Sciences. The work aimed to test whether Large Language Models could support business students without the typical problems of AI-generated misinformation. Four main goals guided the research: building a locally running LLM system, preventing the AI from making up information, checking if standard school computers could handle it, and creating a foundation for future educational AI research. The project used an iterative development approach. A React frontend connected to a FastAPI backend, with Ollama's phi4 model providing the AI capabilities. The theoretical background covered LLM history, vector embeddings, Retrieval-Augmented Generation, and prompt engineering. Arctic Valley's teaching materials were broken into searchable chunks and stored in a Chroma vector database. The finished system met its technical goals. Response times stayed under three seconds, with streaming text making interactions feel faster. Memory use remained around 2GB, confirming the system runs on standard educational hardware. The RAG approach successfully limited responses to information from Arctic Valley selected materials, with the AI clearly stating when it lacked knowledge rather than inventing answers. The modular architecture lets other educational programs to adapt the system. While student testing remains a future step, the technical foundation shows that locally deployed AI can support business education without requiring expensive infrastructure or cloud services.
Source: Theseus (Ammattikorkeakoulujen)
Author: Makarov, Nikita
Language: English
Subfield: Education
Field: Social Sciences
Domain: Social Sciences
SDG: Quality education
Open Access status: green
OpenAlex ID: W7055984630