Research
The research the rest comes from
A PhD at Sapienza's Perception and Intelligence Lab, from 2021 to 2024, and a research period at the University of Amsterdam. Anomaly detection, action recognition and procedural learning: understanding what a person is doing by watching them.
10
published papers
533
citations, as of August 2026
3
as first author
1
Best Paper Award
No jargon: I spent years teaching a computer to watch a person at work and notice that they're getting a step wrong, while they're doing it.
It's the same question a shop-floor supervisor asks walking between the stations. Procedo came out of that, and so does the way I approach the problems people bring me: first you look at how the work is really done, then you decide what to automate.
Where that research became a product →Publications
Most recent first. Where a public repository exists, the GitHub stars are shown.
TI-PREGO: Chain of Thought and In-Context Learning for Online Mistake Detection in Procedural Egocentric Videos arXiv ↗
Noticing in real time that somebody is getting a step wrong, watching the video from their point of view.
Compositional Entailment Learning for Hyperbolic Vision-Language Models arXiv ↗
Models that tie images and words together in a geometric space that respects hierarchies.
Contracting Skeletal Kinematics for Human-Related Video Anomaly Detection arXiv ↗
Recognising anomalous behaviour from skeleton movement alone. First author.
PREGO: Online Mistake Detection in Procedural Egocentric Videos arXiv ↗
Detecting mistakes in a procedure while it's being carried out. First author.
Hyp2Nav: Hyperbolic Planning and Curiosity for Crowd Navigation arXiv ↗
Moving a robot through a crowd without bumping into anyone.
Multimodal Motion Conditioned Diffusion Model for Skeleton-Based Video Anomaly Detection arXiv ↗
The most cited and most used of my papers. First author.
Are We Certain It's Anomalous? arXiv ↗
When the model says "this is an anomaly", how much you can believe it. Spotlight presentation.
Best Practices for 2-Body Pose Forecasting arXiv ↗
Predicting how two interacting people will move. Best Paper Award.
A Self-Supervised Algorithm to Detect Signs of Social Isolation in the Elderly
Recognising signs of social isolation in older people without labelled data.
Query-Guided Networks for Few-Shot Fine-Grained Classification and Person Search arXiv ↗
Finding a person again, or telling very similar categories apart, from very few examples.
Why every paper comes out with the code
A paper that comes out with the code can be redone from scratch: anyone downloads the repository, runs the experiments and gets the same numbers. It's how I worked for every publication, and it's the same thing I bring to projects: what I hand over opens, reads and runs.