projects
All projects
6 in total. Click any row to read the case study.
-
Flow-map learning for turbulent fluids
Neural surrogate models for Nek5000-generated CFD trajectories. Curriculum rollout with Sobolev loss and noise injection for long-horizon stability.
-
Unidiary
AI-guided study planner for university students. Where I learned full-stack: Next.js + Supabase + Clerk + Vercel, end to end.
-
Wind Power Forecasting with Graph Neural Networks
Short-term, turbine-level forecasting on the SDWPF dataset. Spatio-temporal product graphs with k-NN, ε-radius, and a wake-aware "dominant direction" graph across vanilla GCN, FastGCN, and Cluster-GCN.
-
Heterogeneous Missing-Data Imputation with HI-VAE
PyTorch reimplementation of HI-VAE (Nazabal et al. 2020) for tabular data with mixed real/count/categorical/ordinal columns, plus a head-to-head benchmark against GAIN, MICE, and mean/mode imputation.
-
Hydrogen Turbine CFD at E2 Power Systems
A 3-month engagement with E2 Power Systems — a TU Delft hydrogen-turbine startup — establishing a proof of concept for an axial combustion turbine. ANSYS Fluent and CFX cross-validation, blade-shape optimization, and SolidWorks integration of the rotor and fuel-injection components.
-
Pedestrian Classifier for Intelligent Vehicles
A 3-month TU Delft project on real-time pedestrian detection, tracking, and motion planning for intelligent vehicles. CNN classifier with HOG/SVM fallback, Kalman + particle filters for tracking, 2D-to-3D bounding-box projection, and a Best-First-Search planner for constrained urban maneuvers.