M.Sc. student in AI & Robotics @ Sapienza University of Rome — I build research-grade, production-ready ML systems in LLM fine-tuning, knowledge distillation, and self-supervised computer vision.
🏆 1st place internationally for autonomous-flight software design (CSU California UASC) 🔬 Trained models on the Leonardo HPC cluster (CINECA) — 300B → 1.5B distillation
- LLMs — fine-tuning, knowledge distillation, quantization (LoRA/QLoRA), RAG systems
- Computer Vision — self-supervised learning (JEPA/LeJEPA), monocular depth estimation
- Autonomous systems — flight software, reinforcement learning for driving
- HPC — large-scale training with Slurm on the Leonardo cluster
| Project | What it does | Stack |
|---|---|---|
| lejepa-monocular-depth-estimation | Reproducible pipeline evaluating LeJEPA self-supervised representations for monocular depth. Up to 11.6% RMSE reduction on KITTI. | PyTorch, SSL |
| SentenceSplitter | Token-level sentence boundary detection with LLM embeddings + Mixture-of-Experts + multi-scale CNN. | Python, LLMs, MoE |
| HandPointer | Hands-free computer control — real-time hand tracking → pointer + click gestures. | Python, MediaPipe, OpenCV |
LoRA/QLoRA · RAG · Knowledge Distillation · Self-Supervised Learning · Slurm/HPC · CUDA · Weights & Biases
