Sérgio Jesus is the winner of the 4th edition of the “Vencer o Adamastor” Award, which recognises innovative work by young scientists in the fields of electrical engineering, computing, and related areas.
Sérgio Jesus, who earned his PhD in Computer Science from the Faculty of Sciences of the University of Porto (FCUP), won the 4th edition of the “Vencer o Adamastor” Award with his thesis, “Robust Evaluation of Fairness and Explainability in Artificial Intelligence”. The award, worth €20,000, recognises the work of young scientists in the field of information and communication technologies. The award ceremony takes place this Monday, July 6, at 5:00 PM, in the Ferreira da Silva Auditorium at FCUP, and will be attended by the President of the Portuguese Republic, António José Seguro.
The aim of this work is to make AI fairer, because although it helps accelerate decision-making, it can also lead to hasty conclusions that are not always the most accurate. Today, artificial intelligence systems are already deciding who receives a loan, who passes a job screening process, and who is flagged as suspicious. These systems learn from the data used to train them, including the biases contained in those data. They then amplify those biases, as each biased decision can influence future decisions and create injustices. An unfair system can have a very high cost for many individuals and for society as a whole.
With this in mind, Sérgio Jesus, an employee of Feedzai—the data science company that provided him with access to “data, models, and resources” for his PhD research—developed rigorous methods to expose these biases.
He created a methodology inspired by clinical trials and recommender system practices to test the effects of AI and explanations on experts’ decision-making. Rather than relying on simplified automatic metrics, he conducted large-scale experiments with professional fraud analysts.
“My goal was to carry out a more robust evaluation of artificial intelligence systems used in the real world, particularly regarding fairness and explainability,” he explains.
The award-winning research demonstrates that, even when designed with the best intentions and under human supervision, AI can become a machine that jumps to conclusions, amplifying biases and leading to more errors.
Regarding fairness, instead of attempting to remove the biases present in real-world data, using fraud detection as a case study, the young researcher amplified them in order to understand their impact on the models. It was then that he discovered biases suggesting higher fraud rates among older age groups.
Developing resources to promote fairer AI
Sérgio Jesus’s work led to the creation and open-source release of the largest dataset to date for bias research: Bank Account Fraud (BAF). BAF is the first public, realistic, and large-scale tabular dataset designed to study fairness and robustness in AI models under different biased data conditions.
His research concluded that not all approaches are suitable for all types of data bias, reinforcing the importance of robust evaluation practices in scenarios where models may generate discrimination.
A new version of the open-source library Aequitas, called Aequitas Flow, was also developed. It is a comprehensive framework for the systematic evaluation of fairness in AI models. These contributions have been incorporated into the Feedzai Fairness Suite software and made available worldwide by Feedzai. The two open-source resources, BAF and Aequitas, have attracted significant interest, with more than 250,000 downloads. The hope is that they will help researchers develop AI systems that avoid drawing hasty conclusions.
This work, defended at FCUP on September 12, 2025, represents an effort to bridge scientific theory and industrial practice, equipping the community with the necessary tools to develop and deploy AI systems based on ethical and responsible principles.
For Sérgio Jesus, this distinction is “an honour” and a recognition of what was truly a team effort. “It makes me think of all the people who helped me along the way, especially my supervisors and colleagues, who were essential in bringing this thesis to a successful conclusion,” says the FCUP alumnus.
The topic of the research emerged under the “strong influence” of Pedro Saleiro, Chief AI Officer at Opnova, and Pedro Bizarro, co-founder of Feedzai, both specialists in Responsible AI and involved in initiatives such as the Center for Responsible AI.
“We focused on evaluation because we wanted standards in this field to move beyond mere regulatory obligations,” he explains.
Among the supervisors of the award-winning thesis are FCUP faculty member Rita Ribeiro, João Gama, professor at the Faculty of Economics of the University of Porto and researcher at INESC TEC, and Pedro Saleiro.
The Vencer o Adamastor Award is an initiative of the Institute for Systems and Computer Engineering (INESC), in partnership with the newspaper PÚBLICO, and recognises innovative work by young scientists in the field of information and communication technologies that demonstrates not only scientific excellence but also strong potential for economic development.