AI & Data Operations
I build reliable AI-driven data workflows for business systems and quality control. My work automates multilingual processes across 10+ EU markets, stabilizes vendor and payment operations, and integrates computer vision for early anomaly detection — helping teams make decisions with clarity, not chaos.
Projects & Research
Things I’ve built, shipped, or researched recently. Model alignment, medical vision, and data operations at scale.
Tomato Leaf Health Diagnostics (PlantVillage)
Built an end-to-end vision system for tomato leaf disease detection. Designed a reproducible pipeline for data download, label normalization, and deterministic splits. Trained a ResNet18 classifier (PyTorch) on Snellius GPU achieving ~96% test accuracy.
- Deployed via Streamlit (web app) and FastAPI (JSON endpoint)
- Supports real-image uploads with confidence-based feedback
- Next step: in-the-wild fine-tuning for natural lighting data
Python · PyTorch · Streamlit · FastAPI · GPU (Snellius)
LLaMA 2 7B Human Preference Alignment
Compared In-Context Learning (ICL), QLoRA, and Direct Preference Optimization (DPO) to analyze how fine-tuning affects alignment with human preference. Research collaboration within the UvA NLP-2 course.
Python · Transformers · RLHF-style tuning
Medical Image Segmentation (MedSAM vs UniverSeg)
Evaluated generalization in medical segmentation using MedSAM and UniverSeg. Focused on cross-domain performance, Dice score (0.97), and robustness to imaging variation.
Python · PyTorch · medical imaging · Dice 0.97
Multilingual DataOps: Translation Pipeline & Invoice Tracking
Supported finance & procurement flows across 10+ EU markets: stabilized vendor onboarding, mapped region codes, tracked invoice exceptions, and ensured timely payments during entity transitions.
SAP Hybris · data validation · process automation · stakeholder ops
Experience
Work that mixes technical detail with operational impact.
IT Digital & Customer Care Intern
I support finance, procurement, and customer care flows across 10+ EU markets. I keep vendor onboarding and invoice payments stable during an entity/legal change, map region codes, track invoice exceptions, coordinate with Finance / Procurement / Legal, and make sure critical services continue to get paid.
SAP · process automation · multilingual ops · stakeholder alignment
Medical Image Segmentation Research
I benchmark MedSAM and UniverSeg for cross-domain medical segmentation. Focus: model generalization, Dice score stability across imaging conditions, and how to adapt segmentation to unseen data with minimal manual labeling.
PyTorch · computer vision · model evaluation · Dice 0.97
Skills
Technical depth + operational reality. Scroll sideways ↓
Education
MSc Artificial Intelligence
Focus on model alignment (ICL, QLoRA, DPO), RL-style preference optimization, and medical computer vision. Includes work on multilingual, real-world operational systems.
MSc Data Science
Statistical modeling, machine learning, and data interpretation for decision-making in real business settings.
BSc Economics and Business Economics
Focused on econometrics, optimization, and data-driven strategy. Explored how quantitative methods and AI can inform business and policy decisions.