A PiktID product · launched March 2026
Fashion visuals at scale with AI
SaaS for fashion e-commerce: model swap, flat-lay to on-model conversion, and AI identity management. Designed for high-volume batch processing with API integration.
Brands
50+

A collection of my professional experience, research contributions, teaching, and the projects I build on my own time.
Current professional position
Architecting and building from scratch the full GenAI stack powering Studio and On-Model. Full-pipeline ownership: research and development of generative models, backend and API design, cloud infrastructure, and the web applications running on top.
The two GenAI products I've architected and shipped as CTO.
A PiktID product · launched March 2026
Fashion visuals at scale with AI
SaaS for fashion e-commerce: model swap, flat-lay to on-model conversion, and AI identity management. Designed for high-volume batch processing with API integration.
Brands
50+
A PiktID product · launched May 2025
AI image editing platform for human identities
Comprehensive image editing platform featuring identity anonymization, facial expression editing, face swap, full-body and background generation, and image upscaling.
Registered users
300K+
Related reading: what an AI image metadata check actually proves, on the EU AI Act marking work behind On-Model's free image checker.
Academic publications and research contributions
Mutual Information Estimation via f-Divergence and Data Derangements. Plus prior work in IEEE Transactions on Neural Networks, Robotics, and Communications.
Educational experience and knowledge sharing
Four years as Teaching Assistant at the University of Klagenfurt (Machine Learning, Signal Processing, Communications Engineering). NeurIPS reviewer since 2025. Mentor at Lead The Future for Italian STEM students since 2024.
Things I build for fun, and the code behind the papers
A running dashboard that answers one question: am I getting fitter?
Nine panels over my own training data, from Strava or from local .fit and .gpx files.
An arena for judging generative image models against each other.
Side-by-side picks from several raters, ranked with a Bradley-Terry model.
Reference implementations of the papers.
fDIME from NeurIPS 2024, plus CODINE, CORTICAL, MIND and capacity-approaching autoencoders.
I'm always open to discussing new opportunities, collaborations, and interesting projects.
Get In Touch