Projects & Research
A published paper, merged open-source work and award-winning hackathon builds.
- Co-authored a published research paper with one teammate and presented it to the NTT Data CEO and a panel of academic judges.
- Fine-tuned BERTimbau (a Brazilian Portuguese BERT model) in PyTorch to 0.948 macro-F1 on a 6,324-article held-out test set, up from 0.926 for the best classical baseline (TF-IDF + MLP).
- Benchmarked MLP, SVM and Logistic Regression classifiers on TF-IDF features over 63,000+ articles, tuned with 5-fold cross-validated grid search; explained predictions with LIME and permutation importance, and clustered articles semantically using embeddings, K-Means and PCA.
Projects
Open source and hackathons
- Fixed a UI scaling bug, merged by the maintainers.
- Implemented rotate, scale and mirror for bounding boxes on polygon and polyline tracks, with unit tests and Cypress end-to-end tests.
JunctionX Lisbon 2025
Invasive Species Predictor1st Place
Oct 2025
- Won first place in a team of three with a tool for Albatroz Engenharia that flags high-risk invasive plant-species expansion zones from satellite imagery.
- Led the ML component: collected and cleaned the geospatial and satellite data and trained a YOLOv8 segmentation model. The pipeline also visualizes before-and-after scenarios.
HackEurope Dublin
Unscrolling AI Assistant
Feb 2026
- In a team of three, built the backend of a personalized briefing assistant that scrapes and ranks daily updates on user-selected topics, summarizes them with multiple LLMs and turns them into podcast-style audio via text-to-speech. Deployed on Railway.
Junction Finland
BetterMeeting AR
Nov 2025
- In a team of three, built an AR meeting assistant for Snapchat Spectacles: live subtitles with real-time translation, LLM-generated task lists and meeting summaries, and personalized text-to-speech feedback with ElevenLabs.