MSc Student in Computer Science at EPFL

Transformer-based ECoG decoders improved with self-supervised pretraining and vector quantization: +0.25 R² over the supervised session baseline, and R² > 0.4 with only 1% labeled data.

A physically-based ray tracing engine in C++ featuring multiple importance sampling, equi-angular sampling, custom material creation, and volume rendering. Finished among the runners-up of the final project competition, with a jury composed of Prof. Wenzel and engineers from Google, NVIDIA, and Meta.

Opera music analysis and boundary detection using feature alignment, novelty-based segmentation and other processing methods. PyQt6 desktop app (Windows + macOS) with interactive segmentation and real-time audio feature visualization.