AI Research Scientist · Machine Learning Engineer

Hello, I'mGabriele Lozupone

I take deep learning from research to software people use.

PhD in Artificial Intelligence and postdoctoral researcher. I develop generative models, representation learning, and computer vision methods, publish them in peer-reviewed venues, and build the systems that run them: from 3D medical imaging pipelines to an open-source AI photo editor and a mobile app used by more than 1,000 people.

  • Generative & diffusion models
  • Representation learning
  • ML systems & inference
See my work Get in touch View CV

About Me

Research depth, engineering experience, and measurable results

I am an AI researcher and engineer. I take deep learning projects from the first hypothesis through model design, training at scale, and evaluation, to software that other people can install and use.

My main areas are generative AI, foundation models, representation learning, computer vision, and explainable AI.

I developed a latent diffusion framework that made unsupervised learning on 3D medical images up to 20× faster and achieved 90% ROC-AUC in Alzheimer's disease analysis.

My postdoctoral research applies these methods to longitudinal MRI to model how neurodegenerative diseases change over time. Alongside it, I build open-source tools, most recently Kokon, a local AI photo editor.

I believe AI is expanding what we can understand, simulate, and build, making it possible to address problems that once seemed beyond our reach. Contributing to that progress, especially in medicine, is what drives my work and keeps me curious about what comes next.

Research & Engineering

What I bring to a research group or an engineering team

Research

Designing methods, running rigorous experiments, and communicating the results.

  • Diffusion and latent generative models for 3D data (LDAE, Medical Image Analysis 2026)
  • Representation learning, foundation models, and explainable AI, including voxel-level evidence maps evaluated with radiologists (AXIAL)
  • Experiments at scale: multi-GPU and multi-node training on HPC clusters with SLURM, tracked and reproducible
  • International collaboration with Radboud University Medical Center, talks at ESNR and ICPR, and a university seminar on diffusion models
Publications Google Scholar

Engineering

Turning models into fast, reliable software that people install and use.

  • Kokon: an open-source desktop app for local AI image editing, built with Rust, Tauri, React, and TypeScript on top of stable-diffusion.cpp
  • Inference on consumer hardware: quantized models, automatic VRAM planning, and CUDA, Vulkan, and Metal back ends
  • Upstream open source: MoSR upscaler support merged into stable-diffusion.cpp (C++)
  • Agent tooling: an MCP server that lets AI agents call local image tools, published on npm and the official MCP Registry
  • Product delivery: Unicas2Go, a Flutter and Firebase app on the App Store and Google Play with more than 1,000 members
kokon.app stable-diffusion.cpp PR GitHub

Tools I use

ML & research

  • Python
  • PyTorch
  • Hugging Face
  • CUDA
  • Weights & Biases
  • SLURM / HPC
  • Docker

Software

  • C++
  • Rust
  • TypeScript
  • React
  • Tauri
  • Flutter
  • Firebase
  • Git
  • Linux

AI workflows

  • Claude Code
  • Codex
  • AI agents & MCP
  • RAG
  • Local LLMs (vLLM, LM Studio)
  • Cloud LLM APIs
Fully open source · AGPL-3.0

Kokon

AI photo editing that runs on your own computer

A desktop app for Windows and macOS that I designed and built. It runs Qwen-Image 2.1 locally through stable-diffusion.cpp, sets itself up in one click, and ships an MCP server so AI agents can use it. No cloud, no account.

Fully open source (AGPL-3.0): the app, its MCP server, the website, and this video, rendered with Remotion from the app's own React components.

RustTauriReactTypeScriptstable-diffusion.cppCUDAVulkanMetal
Source code Watch in 4K on YouTube Video source Download at kokon.app

Career Timeline

Academic training, research appointments, and selected professional milestones

Positions & education

  1. 2026 – present Current

    Postdoctoral Research Fellow in Artificial Intelligence

    University of Cassino and Southern Lazio

    Conducting research on advanced artificial intelligence methods for modeling neurodegenerative disease progression, with a particular focus on Alzheimer's disease and longitudinal neuroimaging.

    • Brain disease progression modeling from longitudinal 3D MRI
    • Foundation models, generative AI, and representation learning
    • Large-scale experimental pipelines and model validation for medical research
  2. 2023 – 2026

    PhD in Artificial Intelligence

    University of Cassino and Southern Lazio

    Completed the doctoral programme and defended a thesis on artificial intelligence methods for neurodegenerative disease analysis from 3D medical imaging.

    • Explainable and generative methods for Alzheimer's disease research
    • Diffusion-based representation learning for 3D MRI
    • Peer-reviewed publications in medical imaging and artificial intelligence
  3. 2025

    Visiting Researcher

    Radboud University Medical Center

    Research stay at Nijmegen, Netherlands, focused on diffusion models for unsupervised representation learning in medical imaging.

    • Developed Latent Diffusion Autoencoders
    • Established international collaborations with domain experts
  4. 2024

    Advanced Course on Data Science & Machine Learning

    ACDL Summer School

    Specialized training in foundation models and diffusion models, focusing on generative AI for medical applications.

    • Foundation and diffusion models specialization
    • Project development in generative medical AI
  5. 2023

    Master's Degree in Computer Engineering

    University of Cassino and Southern Lazio

    Focused on AI and machine learning techniques with university chatbot development using transformer models.

    • Graduated with highest honors (110/110 cum laude)
    • Thesis on university chatbot development using transformers
  6. 2020

    Bachelor's Degree in Computer and Telecommunications Engineering

    University of Cassino and Southern Lazio

    Focused on computer science fundamentals and deep learning applications.

    • Final grade: 107/110
    • Thesis on BCI movement classification using deep learning techniques

Talks, teaching & awards

  • 2025

    2nd Place - IXH25 Italian XRPL Hackathon

    Italian XRPL Hackathon

    Achieved 2nd place in the advanced cryptography track at the IXH25 Italian XRPL Hackathon. Built a functional and innovative solution in 24 hours with a team of 5, stepping outside our comfort zone to tackle challenges in blockchain and cryptography.

  • 2025

    Invited Speaker - AI in Neuroradiology

    European Society of Neuroradiology (ESNR)

    Invited speaker at the ESNR Online Course on Artificial Intelligence, a 1.5-day international course covering AI fundamentals and applications in neuroradiology including image enhancement, segmentation, and decision support systems.

  • 2025

    Guest Lecturer - Diffusion Models Seminar

    University of Cassino and Southern Lazio

    Delivered a 3-hour theoretical and practical seminar on Diffusion Models for the Deep Learning course, bridging mathematical foundations with code implementation.

  • 2024

    Research Conference Speaker

    International Conference on Pattern Recognition (ICPR)

    Presented research on Alzheimer's diagnosis using handwriting analysis with Transformers and CNNs in Kolkata, India.

Publications

Peer-reviewed contributions in medical imaging and artificial intelligence

Featured

Latent Diffusion Autoencoders: Toward Efficient and Meaningful Unsupervised Representation Learning in Medical Imaging

Gabriele Lozupone, Alessandro Bria, Francesco Fontanella, Frederick JA Meijer, Claudio De Stefano, Henkjan Huisman

Medical Image Analysis (2026)

PDF DOI

AXIAL: Attention-based eXplainability for Interpretable Alzheimer's Localized Diagnosis using 2D CNNs on 3D MRI brain scans

Gabriele Lozupone, Alessandro Bria, Francesco Fontanella, Frederick JA Meijer, Claudio De Stefano

BMC Medical Informatics and Decision Making (2026)

PDF DOI

Transformers and CNNs in Neurodiagnostics: Handwriting Analysis for Alzheimer's Diagnosis

Gabriele Lozupone, Emanuele Nardone, Cesare Davide Pace, Tiziana D'Alessandro

International Conference on Pattern Recognition (2025)

DOI

UniCas Research Initiatives in Medicine and Healthcare

Gabriele Lozupone, Marco Cantone, Emanuele Nardone, Cesare Davide Pace, Ciro Russo, Alessandro Bria, Tiziana D'Alessandro, Claudio De Stefano, Francesco Fontanella, Claudio Marrocco and others

Ital-IA (2025)

UniCas for Medicine and Healthcare

Marco Cantone, Svonko Galasso, Gabriele Lozupone, Emanuele Nardone, Cesare Davide Pace, Ciro Russo, Alessandro Bria, Tiziana D'Alessandro, Claudio De Stefano, Francesco Fontanella and others

Ital-IA (2024)

PDF

End-to-End Asbestos Roof Detection on Orthophotos Using Transformer-Based YOLO Deep Neural Network

Cesare Davide Pace, Alessandro Bria, Mariano Focareta, Gabriele Lozupone, Claudio Marrocco, Giuseppe Meoli, Mario Molinara

International Conference on Image Analysis and Processing (2023)

PDF DOI

Projects & Open Source

Open-source research code and products, from medical imaging to local generative AI

Kokon: Local AI Photo Editor

Public

An open-source desktop app for Windows and Mac that edits photos from a plain-language description, fully on the user's computer. Built on Qwen-Image 2.1 and stable-diffusion.cpp, it installs with one click, picks the model and memory settings for the machine automatically, and supports inpainting, outpainting, upscaling, and an MCP server for AI agents.

Generative AIImage EditingDiffusion ModelsDesktop AppMCP
GitHub Website

AXIAL: Attention-based eXplainability for Interpretable Alzheimer's Localized Diagnosis

Public

An explainable deep learning framework for Alzheimer's disease diagnosis from 3D MRI. AXIAL combines 2D convolutional networks with soft attention to learn volumetric representations and generate voxel-level evidence maps, achieving 85.6% accuracy while keeping model decisions clinically interpretable.

Explainable AIMedical ImagingAlzheimer DiseaseAttention Mechanisms
GitHub

LDAE: Latent Diffusion Autoencoder Efficient Unsupervised Learning in Medical Imaging

Public

A latent diffusion framework for efficient unsupervised representation learning in 3D medical imaging. LDAE operates in a compressed latent space, runs up to 20× faster than conventional diffusion autoencoders, reaches 90% ROC-AUC for Alzheimer's disease analysis, and supports anatomically plausible semantic manipulation.

Diffusion ModelsFoundation ModelsAutoencodersMedical ImagingAlzheimer Disease
GitHub

DiffusionModels-Theory2Code

Public

An open-source educational project connecting the mathematical foundations of diffusion models with production-quality PyTorch implementations. It covers DDPMs, DDIMs, classifier guidance, and conditional generation through a progressive theory-to-code curriculum.

Diffusion ModelsEducational ResourceGenerative AIPyTorch
GitHub

Products & Hackathons

Software shipped to real users, and systems built against the clock

Unicas2Go

Unicas2Go

A production mobile platform for sustainable mobility within the University of Cassino community, adopted by more than 1,000 members. I led its end-to-end development across Flutter, Firebase, route-matching logic, institutional identity verification, real-time coordination, and App Store and Play Store delivery.

Mobile DevelopmentFlutterFirebaseFirestoreGoogle Maps APICarpoolingSustainable MobilityCommunity
Manual Play Store App Store
IXH25 - Italian XRPL Hackathon
🥈 2nd Place - Advanced Cryptography Track

IXH25 - Italian XRPL Hackathon

November 2025

Second place in the Advanced Cryptography track at IXH25. In 24 hours, our five-person team designed and delivered a functional privacy-preserving racing system combining homomorphic encryption, distributed key generation, and XRPL-based transactions.

Team:
Emanuele NardoneSara MarchettiRosa FeraEmanuele Vita
BlockchainCryptographyXRPLAIInnovation
View Details GitHub Watch Demo View Pitch

Contact

For research collaborations, speaking invitations, and professional opportunities

Location

Cassino, Italy

Connect

Message copied to clipboard!

AI Task Blaster

AI Task Blaster

Real-world problems are invading from above! Equipped with my Iron "black-box" arm, fire neural networks to solve challenges in medical imaging, object detection, and beyond. Every shot brings AI closer to victory!

Click on me to join the mission!

Player Ready