XILEF

Developed a novel, state-of-the-art, deep learning-based system to predict visual attention in immersive 360° virtual reality videos by combining both visual and audio cues.

This multimodal approach improves prediction accuracy by integrating spatial sound and dynamic visual features, enabling more realistic and adaptive content delivery in VR environments. The work highlights the potential of audio-visual fusion for enhanced user experience in immersive media.

Designed and implemented a drone coordination system using onboard stereo vision and predictive modeling. The multi-drone formation control algorithm enables a team of UAVs to estimate a ball's 3D trajectory and dynamically position themselves to catch it with a shared net.

This project showcased advanced perception, trajectory prediction, and swarm behavior in a physically realistic simulation framework.

Developed a deep learning pipeline for bone segmentation in spinal CT images using an unsupervised approach based on Active Contour Without Edges (ACWE) loss.

The method segments anatomical structures without labeled data, enabling scalable training on large unlabeled CT datasets. Accuracy is improved, showing strong performance in extracting complex bone structures combining energy-based models with deep learning with a semi-supervised approach using limited manual labels.

At age 16, I created a competitive online first-person shooter featuring fully customizable, destructible environments and diverse game modes. Players can alter the map in real time, adding a dynamic layer of strategy and replayability.

The game received multiple awards, including “Best Game” at several conventions, recognized for its innovation, polish, and depth. Designed to support competitive play while enabling creative freedom through user-driven world manipulation.