MSc in Data Science and Advanced Analytics @ NOVA IMS, Lisboa.
I'm building my first data science portfolio, focused on real-world ML, NLP, and LLM applications. Currently working on my thesis in Radiology Report Generation using deep learning and grounding techniques.
| Project | Description | Stack |
|---|---|---|
| Workers' Compensation Claim Classification | Multiclass ML model to predict injury type for 593K+ insurance claims. Weighted ensemble (XGBoost + LightGBM + MLP). Grade: 20/20 | XGBoost LightGBM Optuna SMOTE LIME |
| Tweet Sentiment Analysis | NLP pipeline classifying financial tweets as Bearish / Bullish / Neutral. Best model: RoBERTa — Macro F1: 78.8% | RoBERTa BERT DistilBERT HuggingFace Optuna |
| ABCDEats Customer Segmentation | Customer segmentation for a food delivery platform (~31.7K customers). Best model: Hierarchical Clustering (Ward) — 4 final segments following CRISP-DM | scikit-learn K-Means Hierarchical Clustering pandas SciPy |
| SIEMENS Sales Forecast | 10-month sales forecasting for SIEMENS Smart Infrastructure. Best model: XGBoost + RFE + Optuna — RMSE: 15.54% (below 20% client requirement) following CRISP-DM | XGBoost LightGBM ARIMA SARIMA Optuna |
| Fidelidade MySavings Chatbot | RAG chatbot for Fidelidade insurance agents over product documentation. 100% accuracy on client Q&A validation set. Supports 6 languages | Azure OpenAI GPT-4o-mini RAG Gradio |
Languages: Python · SQL
ML / AI: Scikit-learn · XGBoost · LightGBM · PyTorch · HuggingFace Transformers · Optuna
NLP: NLTK · Gensim · RoBERTa · BERT · DistilBERT
Data: Pandas · NumPy · Matplotlib · Seaborn · Power BI
Tools: Jupyter · FastAPI · Redis · Docker · Git · Linux
MSc Data Science and Advanced Analytics — NOVA IMS (2024–2026) Thesis: Radiology Report Generation with grounding techniques
BSc Information Systems — NOVA IMS (2021–2024)
🏆 2nd place — Dean's Open Innovation Challenge 2025
