Research Vision
My research aims to lower the barriers between domain expertise and responsible AI development. I design methods, workflows, and software systems that enable domain experts to conceptualize, adapt, evaluate, and deploy AI systems that are transparent, controllable, auditable, and aligned with ethical, legal, and organizational requirements.
The long-term goal is certifiable and practically deployable AI: not only better models, but better AI processes that can be verified, audited, governed, and transferred across domains.
Human-centered AI
Teacher-in-the-loop and human-in-the-loop workflows for accountable AI use instead of fully autonomous replacement.
Responsible deployment
Security concepts, GDPR-aware design, audit trails, risk controls, and safe integration of AI components into institutional systems.
Accessible analytics
No-code and low-code workflows that make model design, evaluation, predictive analytics, and reporting accessible beyond programming-heavy expert communities.
Past and Current Research
Development of a teacher-centered feedback prototype that uses historical submissions and teacher feedback to draft formative feedback for new student work. The results highlight why AI-generated feedback should support teachers rather than replace them.
Development of a baseline protection module for AI components with security controls aligned with institutional requirements, data-protection expectations, and emerging AI governance needs.
Design and evaluation of LLM-based systems for generating language-learning online courses, including human-based evaluation and human-in-the-loop strategies.
Research on transparent models, data quality, imbalanced data, and the limits of relying on aggregate metrics alone when AI predictions affect real contexts.
Quantitative Methods, Predictive Analytics, and Business Intelligence
A core part of my research is the transfer of quantitative methods into robust and understandable analytical environments. This includes forecasting, segmentation, intervention analysis, process analysis, dashboards, and decision support. I am particularly interested in how existing AI components, visual modelling environments, and well-designed methodological scaffolds can make high-quality analysis accessible without reducing scientific rigor.
Software Systems
ML No-Code Editor
Drag-and-drop workflow editor that turns modelled data-processing graphs into executable Python workflows through a FastAPI orchestration service.
Feedback Generator
AI-supported essay feedback system with MLOps infrastructure, Docker, FastAPI, PHP, HTML5, CSS, and React frontends.
Sequencing Generator
Dissertation prototype for automatically selecting and sequencing learning units based on learner performance data, integrated with Moodle.
Learning Analytics Tool
Analytics system for identifying where learners leave self-paced Moodle courses and for evaluating alternative learning paths using large log datasets.
Course Entry Tool
DFKI tool that uses entry testing to help learners assess competencies and skip introductory course sections when learning goals are already met.
Future Research Agenda
- Responsibility-by-design: formalizing verifiable properties, audit trails, risk thresholds, and fallback policies across AI pipelines.
- Technical foundations: reusable no-code workflows with built-in explainability, fairness, robustness, and regulatory alignment.
- Trust and adoption: empirical studies and interface strategies that help users calibrate trust without overreliance or undue skepticism.