AI-driven career path recommendation for software engineering students
Zadeh Hosseini, Melody (2026)
Kandidaatintyö
2026
School of Engineering Science, Tietotekniikka
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe20260620100613
https://urn.fi/URN:NBN:fi-fe20260620100613
Tiivistelmä
Planning a software engineering career is difficult due to the large course selections and specialized paths. At LUT University, the Bachelor's Programme in Software and Systems Engineering offers over one hundred courses totalling 180 ECTS credits, without any tools to match courses to careers. This thesis explores whether large language model (LLM) prompting strategies can generate accurate, career-relevant graduation plans for software engineering students.
This study compares four prompting methods with GPT-5.2. These include a zero-shot baseline, a data-rich method with course context, a step-by-step reasoning approach, and a soft-preference approach. Their performance is tested on two roles using relevance, hallucination, and ECTS completeness. Method A hallucination rates were over 90% making results unreliable.
Methods B and C had 0% hallucinations and similar relevance scores. Method D hallucinations were between 14-25%. Only C produced full ECTS plans, showing the importance of real course data.
This study compares four prompting methods with GPT-5.2. These include a zero-shot baseline, a data-rich method with course context, a step-by-step reasoning approach, and a soft-preference approach. Their performance is tested on two roles using relevance, hallucination, and ECTS completeness. Method A hallucination rates were over 90% making results unreliable.
Methods B and C had 0% hallucinations and similar relevance scores. Method D hallucinations were between 14-25%. Only C produced full ECTS plans, showing the importance of real course data.