Projects & publications

BuddyBack: A Multimodal Smart Posture Correction System

ICMI ’26 - International Conference on Multimodal Interaction, Napoli

A desk-mounted edge device that tracks sitting posture with on-device pose estimation and corrects it through haptic and visual feedback. Published at ICMI ’26; in a 30-person user study every participant reported better postural awareness and none found it intrusive.

Automated Floorplan Extraction and Refinement for Accessible Indoor Navigation

MSc in Computer Science - Sapienza University of Rome

Master’s thesis: a pipeline that turns smartphone photos of evacuation floor plans into clean, editable maps for an indoor navigation system for people with visual and motor disabilities. RT-DETR-L reaches 0.894 mAP@50 and a human-in-the-loop editor replaces manual tracing with quick corrections.

A Sensor Data Collection System for Training a Smart-Parking ML Model

BSc in Computer Science - Sapienza University of Rome

Bachelor’s internship thesis on GeneroCity, a smart-parking app that detects when you leave a parking spot. I designed and built the backend that collects phone sensor data to train its model, and brought the Go codebase under unit tests.

Transfer Learning from Text-Motion Retrieval to Violence Classification

Advanced Machine Learning - Sapienza University of Rome, a.y. 2025/26

Violent actions are rare in motion datasets, so standard classifiers ignore them. We adapt TMR++ text-aligned motion embeddings with first-order MAML and a violence-aware loss: at 30 shots per class the model detects 73.1% of violent actions at a 1% false-positive rate.

Beyond Contrastive Learning: Variational Image–Text Retrieval with RoBERTa and DINOv2

Natural Language Processing · Deep Learning and Applied AI - Sapienza University of Rome, a.y. 2024/25

We adapt VMSST, a variational model from multilingual retrieval, to images and text. It stays within about 6.5 points of a CLIP baseline on MS-COCO, beats it in zero-shot transfer from Flickr30k to COCO, and keeps a structured latent space that a contrastive model does not have.

BuddyBack: Federated Learning for Posture Classification

Training BuddyBack’s posture classifier across several devices without sharing anyone’s data. A FedAvg implementation in PyTorch Lightning over pose features small enough to train on a Raspberry Pi, tested with both even and skewed data splits.

Iris Recognition: Traditional Biometric Pipeline vs Deep Learning

Biometric Systems - Sapienza University of Rome

The classic Daugman pipeline (via the OpenIris library) against ResNet-50 embeddings trained with triplet loss, on the 20,000-image CASIA-Iris-Thousand dataset. The traditional pipeline wins clearly: about 0% EER and 99.7% rank-1 identification.

Pneumonia Detection from Chest X-rays: Classical vs Deep Learning Across Datasets

Fundamentals of Data Science - Sapienza University of Rome

Six classifiers, from k-NN to a fine-tuned ResNet-50, trained on three public chest X-ray datasets and tested on the other two. Deep models win everywhere, but simpler models lose less when the data changes.

AML Challenge: Translating Text Embeddings into Image Embeddings

Advanced Machine Learning - Sapienza University of Rome, a.y. 2025/26

A course competition on model stitching: map embeddings from a text encoder into the space of a separate image encoder. Our VAE with a multi-positive contrastive loss reached an MRR of 0.868 on the public leaderboard, against a baseline of 0.462.

WasaPhoto: A Photo-Sharing Web App

Web and Software Architecture - Sapienza University of Rome

A small social network for photos, built end to end: an OpenAPI-specified REST backend in Go with SQLite, a Vue.js front end, and Docker images for both.