Machine Learning · AI Security · Data Analytics on real-world data

🎓 BSc Applied Computer Science & Data Analytics — UniCa · Jul 2026
🎯 MSc Computer Engineering, Cybersecurity & AI — UniCa · from Sep 2026
📍 Cagliari, Sardinia 🇮🇹 · 🌍 Open to remote & international roles

RAG agent over course PDFs with page-level citations. e5-base embeddings, ChromaDB, Gemini 2.5 Flash.

Multi-stage anomaly-detection pipeline for SETI-like signals on real Sardinia Radio Telescope data — 9-month INAF internship. NCC-max cross-correlation extractor, UMAP + KDE pre-filter, bagged-GMM scoring, memory-mapped training on 1.28M synthetic cadences.

PM10 across Sardinia's monitoring network vs EU limit values — 6M measurements from the EEA archive. One station over the daily limit in five years out of ten; the industrial sites come out cleaner than background. Python pipeline + Power BI dashboard.

wav2vec2 + classification head, served via FastAPI / Docker with CI. Measures the generalisation gap: EER 3.66% in-studio · 8.52% unseen attacks · 29.66% in-the-wild.

CNN deepfake detector under PGD white-box and transfer attacks, hardened with adversarial training (Madry et al., 2018). Recovers ~26 points at ε=0.01 at negligible clean-accuracy cost.

Supervised vs unsupervised NIDS on KDD Cup 1999. RF F1 ≈ 0.999 · One-Class SVM 0.994 · MLP 0.990 · VAE 0.969.

Hands-on SIEM detection engineering on Wazuh. Attacked with Atomic Red Team, wrote custom rules mapped to MITRE ATT&CK for gaps the default ruleset missed — credential access, cron persistence, defense evasion. FIM tuning + rule authoring, validated against each attack.

Credit-card default prediction on UCI dataset. Random Forest: accuracy 0.82, ROC-AUC 0.76. SMOTE lifts minority-class recall 0.37 → 0.47.

Production-style Python data pipeline: uv · pytest · mypy · ruff · GitHub Actions CI. Hot path JIT-compiled with Numba + parallelised with Joblib after Scalene profiling.

Languages

ML & Data

Engineering

Databases & Viz

• Machine Learning & Deep Learning — CNNs, autoencoders, VAEs, GMMs, manifold learning, tree-based & kernel methods
• AI Security — Adversarial robustness, evasion attacks (PGD, transfer), adversarial training defenses
• Blue Team / SOC — SIEM detection engineering on Wazuh, MITRE ATT&CK mapping, adversary emulation with Atomic Red Team
• Data Analytics & BI — Open data pipelines, regulatory thresholds, Power BI dashboards
• Data Science — End-to-end pipelines on real, imbalanced data
• Software Engineering — Modular Python, testing, static typing, CI, FastAPI, Docker, profiling

⭐ Explore my pinned repositories or reach out on LinkedIn.