Farrell Sidadolog

I’m a second-year CS and Data Science student at the University of Toronto, here on a full scholarship from Indonesia’s national merit program. Most of my time goes into machine learning. Last year I spent a few months on a question that turned out to be harder than it looked: can a video transformer outperform a 3D CNN at detecting heart failure from echocardiogram clips, trained on a 6 GB laptop GPU? It can. MViT v2 Small hit ROC-AUC 0.936, 82.5% sensitivity at 90% specificity, with the gap validated by paired bootstrap. Getting there involved a lot of careful data hygiene work — 10,030 videos, deterministic clip windowing, locked thresholds to prevent leakage — before the architecture comparison even started. Before that, I built a global trade network visualisation on BACI data: Louvain community detection across 150+ countries, HHI concentration metrics, interactive Plotly/Dash frontend. I’m drawn to ML problems where the data is messy, the stakes are real, and the baseline is what a domain expert currently does manually. If something here is relevant to you: f.sidadolog@mail.utoronto.ca.