At CESI Lineact in Nice, I’m currently conducting research that sits at the intersection of motion generation and few-shot learning. The aim is to enrich datasets for Human Action Recognition by generating synthetic motions conditioned on textual prompts, a combination that’s still largely unexplored in the literature.
I'm working on a system that integrates generative AI and motion analysis to bridge the gap between simulated and real-world behavior. The work involves designing smart data augmentation pipelines and parameter efficient fine-tuning (PEFT) with a strong emphasis on coherence and control, especially useful in contexts like digital twins and human-centric simulations.
Beyond the technical side, this experience is helping me grow as a researcher: preparing a first-author publication, presenting my ideas, and refining my understanding of how learning-based systems can be used to align simulation and reality.
Motion generation
Few-Shot Learning
Mixture Of Experts
Parameter Efficient Fine-Tuning (PEFT)
Human Action Recognition
📍 Nice - Côte d’Azur - France
Apr 2025 - Present
During my time at Fondazione Bruno Kessler, I collaborated with the "Data Science for Health Care" unit on a research project centered on dimension reduction techniques applied to biomedical datasets, particularly gene expression data. The ultimate goal was to support early disease detection, including complex conditions like lung cancer.
I conducted a thorough benchmarking of both classical and state-of-the-art dimension reduction methods, paying special attention to their ability to preserve meaningful information in high-dimensional, low-sample-size settings (a.k.a. the Big-p, Little-n dilemma). The project also involved evaluating the interpretability of these methods through visual exploration and analysis.
This experience helped me build solid foundations in the field of dimension reduction, and more broadly taught me how to critically approach scientific literature and technical documentation.
Machine learning
Dimensionality reduction
Scientific documentation
Python
📍 Trento, Italy
Mar 2023 - Jul 2023
Achieved first place in the highly competitive 2024 Industrial AI Challenge, an initiative aimed at solving real-world industry challenges through innovative AI-driven solutions. Over the course of 11 weeks, our team developed a multi-stage optimization scheduling solution utilizing a variety of techniques such as Integer Constraint Programming and Genetic Search to address complex scheduling problems.
As part of the challenge, we had the privilege of collaborating with LeMur, gaining hands-on experience with industry-specific constraints and objectives. This collaboration emphasized the importance of effectively understanding and addressing client needs, bridging the gap between technical solutions and real-world requirements. The competition provided an exceptional opportunity to work on practical problems under real-world conditions, requiring a blend of technical expertise, teamwork, and creativity.
The Industrial AI Challenge not only garners the attention of local industries but also attracts major players such as Terna (our competitor this year), Pirelli, and Melinda in past editions. This makes the competition a truly unique and elite opportunity for Italian AI students to engage with cutting-edge industrial challenges.
Job Shop Problem (JSP)
Genetic optimization
Scheduling
Challenge
📍 Trento, Italy
sep 2024 - dec 2024
AI re-ontologizing power is something so exciting and sometimes frightening at the same time it caught my eye just as the "Chat-GPT revolution" came. At the time i told myself I wanted to be part of that, so here I'm trying to do my best while diving into such a thrilling field.
As someone once said, "with great power comes great responsibility" 🕷 🕷 🕷.
That's why I believe as engenieers we often have an under estimated responsibility in shaping peoples' life
trust
commitment
respect
transparency
support
