Projects
EEG-Based Recommender System
Pontifical University of Chile · Recommender Systems Lab
Built a recommender system driven by electroencephalography (EEG) brain signals. Used the DEAP dataset (32 subjects, 32 EEG channels, 128 Hz) to generate topographic brain maps via the MNE library. Fed EEG-derived features into a Transformer encoder with multi-head attention to classify video preferences from brain activity. Constructed user-item interaction matrices and applied collaborative filtering to predict content recommendations. Evaluated with MAP, nDCG, and precision@k.
![[Poster]](https://github.com/Iruretagoyena/RecommenderSystemEEGSignals/blob/main/Poster.png){kind=link}
Ocean Drifter — Computational Fluid Dynamics & 3D Printing
Pontifical University of Chile · Fluid Mechanics
Designed and 3D-printed a Lagrangian ocean drifter for measuring surface currents. Modeled hydrodynamic drag and vortex shedding using computational fluid dynamics simulations. Fabricated custom components (blades, cylinder, housing) via FDM 3D printing, iterated on geometries to minimize drift error. Validated against in-situ oceanographic measurements.
Deep Learning for Human Movement Recognition
Twenty Billion Neurons · Company advised by Yoshua Bengio
Worked on deep learning technologies for real-world video understanding and human movement recognition. Developed and deployed production code to analyze video streams, utilizing custom lightweight 3D-CNN architectures to extract spatial-temporal meaning and classify gestures in real-time. TwentyBN authored the foundational "Something-Something" and "Jester" datasets for video understanding. Company acquired by Qualcomm.
ConvGAN Research — Deep Learning for Image Generation
University of Texas at Austin · Wireless Networking and Communications Group
Conducted deep learning research on convolutional generative adversarial networks (ConvGANs) under Alex Dimakis. Implemented and trained GAN architectures in PyTorch/TensorFlow for image generation and reconstruction tasks. Explored latent space representations and adversarial training stability.
