Projects

StancePrediction

More specifically, "Implicit Stance Prediction on Political Debates using Speech features". The task was addresed with different models: BART, BERT, DistilBERT, wav2vec2 and MulT; the best of which reached an accuracy of 94.2% on the test set.

POS Tagging

Developed a series of neural architectures to approach the "POS tagging" NLP task. Evaluated on the test set and obtained an accuracy of 92.89% and a F1 score of 85.8%.

Fact checking

Trained and analyzed 7 neural architectures to approach the "Fact checking" NLP task with attention mechaninsm, obtaining a standard test accuracy of 78% and a "voting" test accuracy of 79%.

AstroNet

Series of Deep Learning models to detect and classify astronomical sources from radio images. In particular, it was used an Object Detection approach by implementing YOLOv4 from scratch, and a Image Segmentation approach by implementing U-Net also from scratch, the latter achieving an accuracy of 97.3%.

Freebye

App to incentivize circular economy and waste reduction. Developed with Flutter and presented at the Google Solution Challenge 2021.

AshtonTablutAgent

Java-based intelligent agent able to play the game of "Tablut" with the MinMax algorithm. Developed for the "Tablut Students Challenge 2021".

Todaily

Android application for creating and organize personal daily tasks. Using the Java language and the Android Studio development's environment.

Corinne-2

Python tool to manage Choreography Automata. Presented for my Bachelor's thesis in Computer Science and subject of the paper "Corinne, a tool for Choreography Automata" in which Im an author, presented at FACS 2021

Continuous Bezier Curves

MATLAB application that allows the user to view, model and interrogate one or more Bézier curves.

Personal website

The website you are visiting right now. Inspired by the Windows 95 style, as a shy and subtle rebellion against the bland and cold modern web design. Or just because I thought it looked cool.