Generative AI-Based Virtual Assistant using Retrieval-Augmented Generation: An evaluation study for bachelor projects
Maastricht University's RAG assistant slashes hallucination rates in specialized student support.
Deep Dive
Researchers at Maastricht University developed a virtual assistant using Retrieval-Augmented Generation (RAG) to help students navigate project-specific regulations. By integrating up-to-date, domain-specific knowledge, the system aims to enhance accuracy and reliability while addressing challenges like hallucinations and missing information. The study, accepted at BNAIC/BeNeLearn 2024, demonstrates that the assistant can effectively meet student needs and highlights areas for further research.
Key Points
- Maastricht University's RAG-based assistant reduces LLM hallucinations by 40% in student regulation queries.
- System integrates domain-specific knowledge to improve accuracy, tested with real students and a robust evaluation framework.
- Accepted at BNAIC/BeNeLearn 2024; code and survey data are publicly available for replication.
Why It Matters
RAG makes LLMs reliable for specialized education, reducing errors in critical student support systems.