Power Grid Load Forecasting
A data-driven thesis modelling Swiss electricity load with time-series methods and machine learning. Forecasts incorporate seasonality and exogenous weather data, with rigorous diagnostics and evaluation.
PROJECTS
Back to homeSOC Log Anomaly Detection
Customized and deployed an in-house LLM for real-time security-log analysis during a SecuLabs internship. The work included prompt and output normalization, Phi-4 severity ratings, and a Streamlit monitoring dashboard.
Monocular Depth Estimation
Investigated lightweight global-context augmentation and geometric self-consistency regularization for monocular depth estimation, improving contextual reasoning while preserving geometric consistency.
FastGATr Optimisation
Reimplemented and optimized six core Geometric Algebra Transformer kernels in C++. Sparse specialisation, memory-layout improvements, and ARM NEON vectorization substantially reduce runtime on Apple silicon.
Swissgrid: PV Installation-Angle Prediction
Winning Energy Data Hackdays project for estimating PV installation angles across Switzerland.
UR5 Conditional Flow
Robot movement estimation using conditional flow matching for UR5 trajectories.
PINN Darcy Flow
Physics-informed neural networks applied to Darcy flow below a water reservoir.
Industrial VAE
Learning compact representations of industrial imagery with a convolutional variational autoencoder.