The goal of this thesis is to design and prototype a serious game that simulates stakeholder engagement challenges in MLOps lifecycles. The game will make use of LLM-based agents to represent typical stakeholder roles (e.g., data scientists, ML engineers, domain experts, operations managers) and allow players to interact with them in different lifecycle stages. The aim is to both explore the dynamics of stakeholder collaboration in MLOps and provide an educational tool for students, practitioners, and researchers to better understand the complexities of human and organizational factors in MLOps.
Thesis Type |
|
Status |
Running |
Presentation room |
Seminar room I5 6202 |
Supervisor(s) |
Stefan Decker René Reiners |
Advisor(s) |
Michal Slupczynski |
Contact |
slupczynski@dbis.rwth-aachen.de |
Background and Related Work
MLOps extends DevOps principles to machine learning, ensuring reproducibility, scalability, and automation of ML models. While prior work has mainly addressed technical challenges (e.g., CI/CD for ML, model monitoring), stakeholder engagement remains underexplored.
Earlier research has highlighted the need to identify stakeholders, their roles, and their information needs in MLOps. Serious games have been successfully applied in software engineering and requirements engineering to simulate collaborative decision-making and raise awareness of non-technical challenges. At the same time, LLMs can act as role-playing agents, making them suitable to simulate stakeholder perspectives and communication styles.
This thesis combines these strands by developing a serious game prototype where stakeholder roles are represented by LLM agents, offering a novel way to teach and explore engagement in sensor-based MLOps lifecycles.
Scope and Deliverables
- A game prototype (web or lightweight interactive simulation) where users interact with LLM-based agents representing stakeholders. The serious game will simulate decision-making scenarios across different stages of the MLOps lifecycle. The player takes on the role of one stakeholder (e.g., data scientist) and interacts with LLM-driven agents representing the other stakeholders (e.g., ML engineer, domain expert, operations manager).
- A set of predefined scenarios (e.g., model monitoring, retraining decision, data quality issue) reflecting MLOps lifecycle challenges.
- An evaluation study (small-scale, e.g., with students or peers) to assess usability and educational value.
- A short reflection on limitations and future improvements, e.g., scaling to more complex stakeholder interactions.
Expected Contribution
The thesis will deliver a proof-of-concept serious game that demonstrates how LLM-based agents can simulate stakeholder roles in MLOps. This contributes to both educational technology in AI/ML contexts and novel methods for studying stakeholder engagement in MLOps lifecycles.
- Basic programming skills (Python, JavaScript, or game frameworks).
- Familiarity with machine learning workflows (high-level understanding of MLOps).
- Interest in LLM APIs (e.g., GPT) and conversational agents.
- Optional: Experience with game design or gamification is a plus.