“try to long press the figure 😉”
PhD student in Smart Computing and Artificial Intelligence at the University of Florence, working in collaboration with the University of Siena. Current research focus is fake and synthetic-content detection, especially for images: the study of which representations preserve useful traces of the generation process and how those traces behave across models, datasets, and common image transformations.
An animated terminal screensaver collection and generative ASCII/ANSI art engine featuring native C visualizers (60 FPS), transparent JIT compilation, pure Bash fallback, and macOS anti-sleep protection.
A Python library for extracting internal hidden states and attention maps from Stable Diffusion 3.5 MM-DiT blocks for synthetic image detection, model attribution, and diffusion interpretability.
A micro-LLM (~8.2M–12.1M parameters) trained on medieval Italian literature and Dante Alighieri's works, executing entirely in the browser via ONNX Runtime Web with constrained terza rima decoding.
A Python CLI process wrapper that streams live command output, hardware telemetry (CPU, RAM, GPU, VRAM, Disk), and interactive process controls to a single auto-updating Telegram dashboard.
A conditional autoencoder for exploring chemical representations relevant to the early-stage screening of pharmaceutical contaminants for adsorption by phenyl-functionalized poly(2-oxazoline).
A lightweight Python feed generator that scrapes GitHub Trending repositories across languages and time windows (daily, weekly, monthly) into static Atom XML feeds and a browsable HTML directory.
A study of density estimation from unlabeled samples, combining estimates from maximum-likelihood Gaussian mixture models with the generalization of an artificial neural network.
A PyTorch Generative Adversarial Network trained on synthetic parametric harmonograph and Lissajous curves, investigating adversarial training dynamics, gradient penalty, and Sliced Wasserstein Distance.
An ESANN 2022 comparison of DeepLab V2 and MobileNet approaches for segmenting glomeruli in histological whole-slide images from mouse and rat tissue.
A self-hosted Telegram RSS/Atom aggregator featuring feed auto-discovery, social media resolvers (YouTube, Reddit, X), Telegraph Instant View conversion, extractive TL;DR summaries, and OPML portability.
A visual art collaboration with Gabriel Volpi that recontextualizes convolutional segmentation masks and neural network boundary artifacts from histopathological whole-slide tiles into an abstract digital series.
A web-based inventory and task management application built with PHP, MySQL, and QuaggaJS for client-side EAN barcode scanning and relational storage tracking.
