Development of machine learning models to classify textual content for moderation purposes.In particular, I am responsible for:• creating taxonomies of harmful content (e.g. sexual content, violent content, etc.) divided into different levels of intensity,• building annotated datasets from various sources, leveraging human annotation as well as machine learning models (e.g. transformer, LLMs), and • training a variety of machine learning models to tackle challenges such as textual content moderation, multilinguality and word sense disambiguation, comparing different state-of-the-art approaches.

Research internship on Natural Language Processing in the Aizawa Lab, NII (https://www-al.nii.ac.jp/) under the supervision of Prof. Akiko Aizawa.The aim of the project was to investigate the robustness of recent large language models (LLMs), mainly exploring their potential harmfulness and ways of mitigate them. Through my research, I :• clearly defined the notion of ethics, harms and fairness in the context of natural language processing, exploring current ethical guidelines from NGOs (e.g. UN) and model developers (e.g. OpenAI and Perspective API).• designed an evaluation schema for assessing the harmfulness of texts, taking into account the full complexity of this process, and provided clear annotation guidelines to promote transparency and objectivity in the evaluation.• assessed the robustness of OpenAI LLMs against adversarial prompts trying to circumvent value-alignment effort.

Research project on Natural Language Processing in the NLP Lab, EPFL (https://nlp.epfl.ch/) under the supervision of Prof. Antoine Bosselut.The aim of the project was to create a dataset for Natural Language Inference (NLI) with long-context premises, using crowdsourcing. The focus was primarily on having a challenging dataset that requires commonsense reasoning & multi-sentences reasoning to evaluate the existing SOTA models on it.

Summer internship on Applied Machine Learning at the Machine Learning for Education (ML4ED) Laboratory, EPFL (https://www.epfl.ch/labs/ml4ed/) under the supervision of Prof. Tanja Käser.Time-series analysis of behavioral data to predict students' learning strategies and motivationin an intelligent tutoring system — Lernnavi (https://www.lernnavi.ch/).It resulted in a better understanding of students' learning behaviors through a study conducted in several high schools in Switzerland in collaboration with the University of St-Gallen and Taskbase.

Data & Software Project Manager for the COVID-19 contact tracing center. (60%)Collaboration in the COVID-19 contact tracing center to develop an analytics pipeline and a backward tracing software to understand and prevent chains of contamination, as well as implementing new data collection processes.Main responsible for exploring COVID-19 data with a focus on its evolution and producing analysis to support the decision-making of the senior management and the politics of Vaud state.For the last few months, also responsible of working on the anonymisation of the dataset containing the sensible data related to the Covid-19 of the Vaud state. This, in order to produce a version of the dataset that could be « publicly » accessible for research.

Preparation and supervision of exercise sessions for student support, team manager for students projects and assistance to teachers in various tasks such as proofreading, exam monitoring and correction.Assistant for several editions of the classes:• Introduction to programming [CS-111(i)] with Dr. Jamila Sam• Object-oriented programming [CS-112(i)] with Dr. Jamila Sam• Project oriented programming [COM-112(a)] with Dr. Ronan Boulic• Information, Computation & Communication [CS-119(c)] with Dr. Ronan Boulic & [CS-119(g)] with Dr. Jamila Sam and Dr. Barbara Jobstmann

Implementation of several algorithms of artificial intelligence with a GUI to display them step-by-step, in order to have them be used for educational purposes in the class CS-330 "Intelligence artificielle" given by Prof. Faltings at EPFL.Algorithms :• Graph search (DFS, BFS, iterative DFS, A*)• Constraints satisfaction (Forward check, Backtrack, Backjump, Look-ahead)