Things I have built and studied — at work, at university and on my own.
Automation & data2026
Data Vizz — automated video publishing
Personal project
Data Vizz publishes bar-chart-race videos on YouTube and Facebook. A daily Python pipeline uploads every new video to both platforms with its title, description, tags and a topic thumbnail, and cuts a vertical Short / Reel from it with ffmpeg.
This site: all text kept in one content file, rendered to static pages by a small Node script with no framework or dependencies, and served by Cloudflare Workers, in light and dark themes.
JavaScript
Node.js
HTML
CSS
Cloudflare Workers
Automation & data2025
Purchase-order tracking and reminder reports
SELISE · for a European IT distributor
Maintained Power Automate and Power Query pipelines that turn Dynamics 365 purchase-order data into automated tracking and reminder reports, and traced faults in multi-step refresh chains back to their source data.
Production system simulation and multi-objective optimization
Uppsala University · group project
Modelled an assembly line, justified the warm-up period, identified bottlenecks and interpreted the Pareto-front results of a multi-objective optimization.
FACTS Analyzer
Discrete event simulation
Machine learning2025
Emergency-room waiting-time prediction
Published2nd IEOM World Congress · conference paper
Predicts ER waiting times from 5,000 patient encounters, comparing linear models, tree-based ensembles and support vector machines; a tuned stacked ensemble reached R² = 0.9473.
Breast cancer prediction with ML and deep-learning ensembles
PublishedPLoS One · co-author
Trained and compared KNN, SVM, ANN, Random Forest, XGBoost, AutoML and deep-learning ensembles on original and synthetic datasets, showing how ensembles and synthetic data improve diagnostic accuracy.
Combines fuzzy AHP, Bayesian Belief Networks and machine learning in a user-friendly interface; the model picks the most suitable supplier with over 95% accuracy.
Compares ML classifiers on the UCI Heart Disease data — AutoML came out best — then assesses heart-disease risk for predicted cases with a Bayesian Belief Network.