Julius Gabelmann 🎾

he/himJulius Gabelmann

Junior Researcher

Deutsches Forschungszentrum
für Künstliche Intelligenz

Universität des Saarlandes

Campus D3 2, Room +2.10

66123 Saarbrücken

Germany

About Me

I am a junior researcher in the Neuro-Mechanistic Modeling group, working on applied AI with a focus on agentic systems built on large language models.

In my Master’s thesis at Saarland University, supervised by Prof. Dr. Verena Wolf, I developed ALA (Artificial Learning Assistance), a modular LLM-based tutoring assistant for a university statistics course with more than 100 participants. ALA routes each student query to a specialised module (hints, explanations, evaluation, or general questions) and enforces pedagogical constraints so that it supports learning instead of handing out solutions. I deployed and evaluated it in a live course setting.

Beyond educational AI, I am interested in deep reinforcement learning, multi-agent routing, and retrieval-augmented systems.

Projects

  • preAIrranging – Preparing an AI-Based Rearranging Hub (abat+, DFKI, Saarland University): we use neuro-symbolic approaches to solve the combinatorial optimization problems faced in modern production processes in industrial manufacturing.

Interests

I work on LLM-based agents that are useful in practice and behave predictably: systems that route requests to specialised components, respect explicit constraints, and can be evaluated rigorously. In my freetime, I like tennis and music.
  • Agentic AI Systems
  • LLM-based Tutoring & Educational AI
  • Deep Reinforcement Learning
  • Task-Oriented Dialogue & Multi-Agent Routing
  • Retrieval-Augmented Generation

Education

  • M.Sc. (Data Science and Artificial Intelligence), 2026

    Supervised by Prof. Dr. Verena Wolf at Saarland University

    Artificial Learning Assistance: A Modular Architecture for LLM-Based Educational Chatbots

  • B.Sc. (Data Science and Artificial Intelligence), 2023

    Supervised by Prof. Dr. Bernt Schiele at Saarland University

    Few-Shot Learning with the Help of Self-Supervision