AI/ML Engineer & Data Scientist based in Rome, Italy.
I build end-to-end machine-learning systems, from data pipelines and model evaluation to LLM/RAG applications and reproducible cloud deployments. My current work focuses on reliable AI, agentic systems, information retrieval, and anomaly detection.
- M.Sc. in Computer Engineering, Roma Tre University — 110/110 cum laude
- Former Visiting Research Scholar in Computer Science at The University of Alabama
- First author of Cognitive Firewall and co-author of Agentra, accepted at IEEE ICMLA 2026
- Interested in AI/ML engineering, applied research, LLM safety, RAG, and MLOps roles
- Cognitive Firewall: reduced jailbreak attack success to 2% or below on three of four benchmarks while keeping benign over-refusal at 8%.
- Agentra: helped design a supervisable multi-agent LLM system that raised decision F1 from 0.61 to 0.84, with a 0% projected harmful-action rate on a 120-event corpus.
- Early network intrusion detection: evaluated BiGRU, Transformer, Random Forest, and XGBoost models using only the first packets of network flows, while exposing benchmark leakage through controlled explainability experiments.
- AI4ESOPP: built a Maskable PPO scheduler for energy-aware industrial production that achieved zero weighted tardiness across evaluation episodes.
| Project | What I built | Focus |
|---|---|---|
| Cognitive Firewall | A zero-trust oversight framework that evaluates intent, context, conversation consistency, and output before an LLM responds | LLM safety, evaluation, local LLMs |
| MVTec Anomaly Benchmark | A reproducible benchmark and live Gradio demo comparing five industrial anomaly-detection models across MVTec AD | Computer vision, PyTorch, MLOps |
| Amazon Reviews Sentiment Analysis | A comparative evaluation of rule-based, classical ML, and generative approaches for binary and multiclass sentiment analysis | NLP, scikit-learn, LLM APIs |
| Robust Organ Scheduler | A decision-support prototype using robust min-max optimization for uncertain multi-organ transport scheduling | Optimization, algorithms, Python |
| Recipe Search Engine | A containerized Italian full-text retrieval pipeline with bulk indexing and interactive search | Elasticsearch, Docker, IR |
Languages: Python, Java, C, SQL, Bash
ML & AI: PyTorch, TensorFlow/Keras, scikit-learn, Transformers, reinforcement learning, LLMs, RAG, LangChain, LangGraph
Data & retrieval: pandas, NumPy, Elasticsearch, ETL, knowledge graphs, BM25, nDCG, MAP, MRR
Engineering: Docker, Kubernetes, Git, CI/CD, Linux, REST APIs, AWS, Google Cloud, Azure
Security: intrusion detection, network-flow analysis, anomaly detection, MITRE ATT&CK, NIST CSF
I care about systems that are not only accurate, but also measurable, explainable, and safe to deploy. My work emphasizes controlled evaluation, error analysis, reproducibility, and auditable decision-making.