A Master's student of Artificial Intelligence at Le Mans University, with a software engineering foundation and a researcher's appetite for models that hold up under scrutiny. I build deep learning systems end-to-end — from the data to the benchmark.
I like models that work — that generalize, ship, and survive a proper evaluation.
Before the Master's in AI at Le Mans, I finished a CS degree at Montpellier and spent a summer at LIUM reproducing the state of the art in audio-visual deepfake detection. The common thread is a preference for building from scratch: datasets, training loops, judges, interfaces — the parts that turn a paper into something usable.
I’m currently seeking a long-term role where I can bring value, collaborate effectively, and contribute to meaningful projects.
A learning platform for medical students built on a RAG system that generates yes/no questions using verified knowledge and controlled inaccuracies to encourage critical thinking. I contributed the SLM behind the question-generation pipeline; presented at two conferences including an international venue.
A contextual-bandit pipeline that learns to pick optimal audio prompts for ZipVoice, a one-shot voice cloning model. SONAR embeddings + FAISS search + a deep residual policy trained on CER-based reward — over a 63k-segment French-speech database.
A pedagogical agent for medical education: LLMs & SLMs generate complex yes/no questions, grounded by a RAG layer over a verified medical database. React front-end, Django API. Basis of the Springer publication.
A time-series forecasting service that predicts future GitHub repository activity using a sliding-window approach over six-month intervals. MLflow for tracking, FastAPI + Docker for serving, Streamlit for exploration.
A deep-vision playground: CNN classifiers (binary + multi-class) with transfer learning, GAN image generation, and a VAE with latent-space controls used for colorization, denoising, and generation.
I'm open to internships, research collaborations, or a good conversation about AI, evaluation, or anything adjacent. Fastest way to reach me is email.