Master thesis 2024 at TH Würzburg-Schweinfurt
Case study · RAG Educational AI System
A RAG learning system built from lecture material.
Master thesis with grade 1.0. Later published as an ITHET 2024 paper and awarded the Hans-Wilhelm Renkhoff Award.
In short
Slides in. Learning questions, assessment and progress out.
The core was not a free-form chatbot. It was a small learning flow with source grounding.
Grade 1.0 · ITHET 2024 paper · Renkhoff Award
Retrieval, learning questions, assessment, progress
Research project, not a permanently operated SaaS
System flow
From material to learning questions and progress.
A simple chain: find the source, generate a question, assess the answer, show progress.
01Lecture Slides02Retrieval03Question Generation04Assessment05Progress
Contribution
What I built.
Made lecture material usable as a checkable source for answers and learning questions.
Brought chat, generated questions, assessment and progress into one learning flow.
Built and evaluated the system, then published the work as a paper.
Learnings
Do not make it bigger than it is.
Strong as a documented research and build project. Not a replacement for production metrics.
Sources need to be visible. Otherwise RAG can look more certain than it is.
A learning product needs more than chat: tasks, feedback and progress matter.
Evaluation matters more than a smooth demo.
Proof