Matteo Tafuro

Do artworks influence the way we think about gender? Or do artworks merely reflect common conceptions of gender prevalent in a given era and culture? The goal of this project is twofold. The first goal is to explore whether AI can identify different conceptions of gender (normative and non-normative) in visual artworks. Understanding whether different conceptions of gender in visual artworks can be systematically detected by AI would be a stepping stone to addressing the second goal of this project: to explore whether the diversification of gender we have seen in the social world predates, concurs, or postdates portrayals of gender in visual artworks.

• Led seminars, supervised students and graded assignments/exams. • Developed and maintained Jupyter-notebook course assignments, ensuring reliable virtual environments, consistent formatting and availability of solutions for teaching assistants.

• Played a key role in pivoting the startup’s business model, driving rapid growth and supporting the company’s successful scaling within the first year. • Led the design and deployment of scalable LLM-based solutions to automate logistics operations, working closely with data scientists, domain experts and stakeholders. Leveraged AWS and Azure cloud infrastructures together with RAG techniques to improve domain-specific accuracy, achieving an 88% reduction in manual operations for a major client. • Drove the end-to-end delivery of several PoCs, from defining methods and estimating resources to execution. Each PoC successfully evolved into a prototype, including one that sealed a key partnership with the startup’s anchor client.

• Joined the Cardiovascular Lab of the qurAI research group to work on my Master’s Thesis, as part of the Amsterdam ELLIS honors program. My research was conducted under the co-supervision of Ivana Isgum (University of Amsterdam) and Alison Noble (University of Oxford). • Developed a novel method for obtaining temporally consistent segmentations from sparsely labeled echocardiograms, leveraging image registration to generate pseudo-labels for unlabeled image frames and therefore enabling the supervised training of 3D (2D+time) segmentation models. • Culminated in a publication at MICCAI 2023 and a presentation in Vancouver, Canada, highlighting its role in advancing ultrasound imaging analysis.

• Moderated a roundtable at Toloka’s social event at NeurIPS 2022, facilitating discussion on building bias-robust ML systems, creating representative datasets, and resolving annotator disagreements to support high-quality AI products.

• Served as a Student for a Day (SFaD) mentor at the University of Twente, guiding prospective students through campus visits and accompanying them to bachelor-level lectures, practicals, and workshops to provide first-hand insight into university life.

Summer Job in Portsmouth, UK.