These are the things I have built, from school to today. They run from a dancing robot to a myoelectric hand.
Self Orienting Neck Natural Yarn
2015 → 2017
Won1st · RomeCup 2017
SONNY stands for Self Orienting Neck Natural Yarn. I built it at ITI Archimede, my technical high school in Catania, with Arduino and a Raspberry Pi.
It is a robot that can dance, talk and see.
It was a school project, and it had to work in front of other people, in competitions.
SONNY won RomeCup 2017 and the Italian Robotics Olympics 2017. Before that, I was part of a team that won RomeCup 2015.
Ex Machina Ad Hominem · 3D-printed myoelectric prosthesis
Funding€18k secured
ExMAH, Ex Machina Ad Hominem, is a 3D-printed myoelectric prosthesis. It started at Politecnico di Torino under Prof. Marco Knaflitz, and I co-founded it and led the student team. I was the youngest of the founders.
The team was 10 to 15 students from four departments, and we secured €18k to build it.
On the technical side, muscle signals are picked up by double-differential silver electrodes and an op-amp front end. A 16-bit, 8-channel ADC samples them at 40 kHz and an STM Nucleo microcontroller handles them. Three servomotors move the hand. The project also includes firmware, a GUI, preprocessing and machine learning.
We tested the prosthesis on healthy subjects, with a dataset of our own. It has no force feedback. My own summary of it: failures, satisfactions, a lot learnt.
Automated segmentation from CT slices
Volumes50 CT volumes
This project segments the carotid artery automatically from CT. The input is 50 volumes, each built from stacked CT slices in DICOM format.
It uses two U-Net models with a ResNet-64 backbone, so that the internal and the external carotid are segmented separately. The volumes were split 35 for training, 5 for validation and 10 for testing.
A post-processing step follows, and the results were compared with manual masks.
The pipeline segments both carotids and was checked against manual masks. I report no further metrics here.
Landmarks, a neural network and speech synthesis
Accuracy86% accuracy
The ASL translator turns signs into text and speech. It relies on MediaPipe, which extracts landmarks from video, and a neural network that recognises the sign from those landmarks.
I built a dataset for it: 1500 videos of five seconds each, recorded across different users and conditions.
A GUI ties the pipeline together and adds speech synthesis.
The system reaches 86% accuracy, with a latency of a few seconds.
IoT garden and pool management
StackMQTT + REST
My Garden is an IoT system for managing a garden and a pool. I built it with D. Merlin, A. Ravera and A. Navone.
The hardware is a Raspberry Pi and an Arduino. On the software side there is a microservices server, a catalogue manager that keeps its data in JSON, a Telegram bot and a web app made with Node-RED. The parts talk to each other over MQTT and REST.
The system also takes irrigation decisions automatically, with the aim of saving water.
A working IoT stack, from the hardware to a chat bot and a web app. I have no figures on the water saved.
A dataset for an Expected Goals (xG) algorithm
StackR · Python · MATLAB
Expected Goals, or xG, estimates how likely a shot is to become a goal. This project is the groundwork for such an algorithm: the dataset.
I scraped the data from the web in R, then merged it in Python and MATLAB. After that came feature engineering and shot-density plots.
Football is something I watch, play and now analyse with data.
The outcome is the dataset and its plots, the base on which an xG algorithm can be built.
Prototypes built fast, with teams from several cities
EditionsGeneva · Zurich · Oxford · Turin
Hackahealth is the one I did most: I took part in all of its Geneva and Zurich editions, building prototypes for people with disabilities.
There were others: Oxford Hack, the Turin Google Developer Hack and more.
Prototypes, not products. Each one was built in the time the event allowed.
An STD-awareness web app and startup venture
Ran forTwo years
Bridge was a startup venture: a web app to raise awareness about sexually transmitted diseases. It was born from the SEI Inventor Program.
We built partnerships with Italian Local Health Boards (ASL) and ran a risk survey that used machine learning.
We kept at it for two years.
Bridge closed after two years. The business model was not sustainable, and the team effort was not enough. I say so openly: it is part of what I learnt.
A personal venture, stay tuned
StatusStay tuned
Enosis is a personal venture, and it is still being built. The name is the Greek ἕνωσις, union.
It combines wearable firmware, a mobile app and a backend.
That is all I am saying for now. Stay tuned.
Still being built. Nothing to show yet.