Hi, my name is
I build trustworthy AI systems.
I'm an AI/ML researcher and master's student in Trustworthy and Responsible AI at École Polytechnique (IP Paris), currently exploring geometric deep learning for mechanical design optimization at Michelin.
I'm a master's student in Trustworthy and Responsible AI at École Polytechnique (IP Paris) and an AI/ML researcher with first-author and co-author publications across international conferences and journals. My work spans deep learning, NLP, recommendation systems, physics-informed and geometric deep learning, and optimization.
I've carried out research across academic labs, industry AI teams, and government-backed initiatives — applying machine learning to computational psychology, healthcare and disease surveillance, geospatial and climate modeling, industrial process optimization, and satellite navigation. These days I'm an AI research intern at Michelin, on the Industry 4.0 Exploration team.
Alongside the technical work, I have a sustained interest in AI ethics and responsible deployment — transparency, fairness, and the societal impact of the systems being built.
Technologies I work with regularly:
Raised the best F1-score for text-based depression detection from 93.24% to 98.04% using sentence transformers and Bayesian-optimized classifier ensembles.
A pipeline to subtitle and dub films into English while preserving actors' voices and emotions — speech separation, translation, and voice synthesis.
A PINN framework modeling ODE systems of control and state vectors, achieving new best values for the batch-reactor and tubular-reactor problems.
A hybridized music-recommendation system combining deep contrastive learning with collaborative and content-based filtering — a 24.2% performance gain over baselines.
A deep neural network framework for ODE-based dynamic optimization that outperformed metaheuristic and gradient-based methods, setting a new best solution for the ethanol-production problem.
An NLP-based depression-detection model achieving a 93.24% F1-score on a curated dataset of 989 Reddit posts.
A Random Forest rainfall-classification model on INSAT-3D satellite OLR data, improving the Critical Success Index by 30.2% over existing meteorological methods.
A zero-shot NLP pipeline scoring clinical patient notes against medical rubrics using BioBERT-based QA and NER — no supervised data required.
An agent-based simulation of how political WhatsApp forwards spread, combining social-network graphs with opinion-dynamics modeling.
Exploratory multivariate analysis of 6,497 wines in R — regression, ANOVA, PCA, and factor analysis — with 92.4% colour-classification accuracy.
06. What's Next?
I'm currently open to internship and research opportunities in AI/ML. Whether you have a role in mind, a potential collaboration, or just want to say hi, my inbox is open.
Say Hello