I study how people form beliefs about, rely on, and oversee imperfect AI systems. I am a PhD researcher in Human-Centered AI at TU Delft, working on AI evaluation, human oversight, and repeated human–AI interaction.
My work combines behavioral experiments and statistical modelling with the systems needed to run studies. Before my PhD, I worked in machine learning and enterprise software engineering and co-founded Qaring.
Google Scholar · TU Delft · LinkedIn · Email
I maintain Aligned AI Stack, where I bring together study software, analyses, and experimental tools around belief, reliance, and human oversight of AI. The organization connects the published studies below with ongoing work on how people can assess, question, and correct AI assistance.
- AI evaluation and oversight. Studied how 450 participants used different forms of AI support to report illegal content under systematically varied AI errors. The work examines both decision accuracy and the quality of people's explanations. FAccT 2026
- Belief updating and delegation. Designed a preregistered experiment with 240 participants and 7,200 trials to study how experience with AI carries across tasks and shapes delegation. CHI 2026. Follow-up work separates whether beliefs change from how strongly people update. HCOMP 2026
- Multilingual human–AI interaction. Ran linked co-writing and donation experiments to study how language-dependent AI performance affects reliance and persuasion. CHI 2025
| Project | What is here |
|---|---|
| CHIME | Human annotation tools and analysis for causal model explanations. HCOMP 2022 paper. |
| SECI Persona Studio | A research workspace for comparing source-linked evidence, preserving project decisions, and developing reports. Next.js, FastAPI, Electron, and Ollama. |
| Rift Rewind | A hackathon prototype that turns League of Legends match histories into an interactive narrative. React, behavioral analytics, and an AWS processing pipeline. |
Some study repositories contain participant data or working research material and remain private. Papers above describe the methods and findings.
Preregistered experiments, behavioral measurement, mixed-effects and hurdle models, causal inference, and longitudinal analysis. I build research tools with Python, TypeScript, React, FastAPI, Docker, local language models, and AWS.
I am interested in industry research roles in human–AI interaction, AI evaluation, and oversight.