AI Engineering · Télécom Paris · Institut Polytechnique de Paris
I work at the intersection of machine-learning research and entrepreneurship. From signal processing and computer-vision papers to founding an education company and leading the AI society of Télécom Paris.
I am an engineering student at Télécom Paris, specialising in machine learning, imaging and audio signal processing. In 2026-27 I am on an academic gap year in Shanghai, building an industrial AI & retrieval-augmented-generation system as an AI engineering intern at Pierre Guérin.
Alongside my studies I served as president of Télécom IA, the school's artificial-intelligence society, bringing speakers like Luc Julia and Gael Varoquaux to campus and co-organising research hackathons with IBM, Margo and Qubit Pharmaceuticals. I also founded Choten, an education company, and co-founded the startup Summ'Up.
Selected work from the Télécom Paris image, data and signal track, written with the rigour of papers and built with working code.
Adapting NAF (Neighborhood Attention Filtering) to hyperspectral imagery and training it on frozen DOFA foundation-model features by self-distillation, a fully label-free pipeline compatible with unsupervised spectral unmixing. On an urban EnMAP scene it beats bilinear upsampling of the same features (endmember SAD 0.072 vs 0.099; abundance NMSE 0.143 vs 0.147).
Separating K sources from an under-determined stereo mixture using MDCT sparsity, magnitude-weighted direction histograms and MMSE reconstruction.
β-divergence multiplicative updates derived and implemented from scratch; pitch-level analysis of a piano scale and Wiener-style source reconstruction.
EfficientNet-B2 deep features fused with 25 handcrafted morphological descriptors over 13 classes; class-balanced focal loss and test-time augmentation.
TF-IDF vectorisation + MiniBatchKMeans bucketing over hundreds of thousands of phrases to trace how memes appear, mutate and spread through the press and blogs.
Reconstruction of the Stanford Bunny (36k points) with adaptive alpha selection, from raw point cloud to watertight STL meshes.
A critical review of time-delay DNNs as UBMs in i-vector speaker-recognition systems: methodology, NIST SRE10 results, strengths and limitations.
A close reading of the TD-PSOLA framework for text-to-speech: pitch-mark analysis, prosodic modification and overlap-add synthesis, worked through on real diphones.
Aliasing under naive downsampling versus low-pass-filtered decimation, and an analysis of hue/saturation behaviour in colour space.
Télécom IA is the AI society of Télécom Paris and a co-founder of Unaite, the federation of AI societies across France's top schools. As president, I led the team behind its flagship events.
70 selected students, three days on molecular-toxicity prediction for drug discovery. Covered by Le Monde Informatique.
An evening with the co-creator of Siri, former Apple VP and Chief Scientific Officer of Renault, co-hosted with the Télécom Paris alumni network.
Conference with the Inria research director behind one of the world's most-used machine-learning libraries.
A hands-on session on tabular foundation models (TabPFN, TabICL), co-run with Onepoint: how transformers pre-trained on synthetic tables now beat gradient boosting on small tabular datasets, tried out on real data.
A session with Converteo on designing agentic AI systems: how to architect tool-using, multi-step LLM agents that are reliable enough to ship.
Student-led sessions on StyleGAN2 and AI fundamentals, plus partnering on HackEurope, Europe's largest student hackathon.
Founder and president of Choten Académie, an education company running intensive mathematics and physics programmes for French high-school and classes préparatoires students, with a mentoring method built by graduates of France's top-5 engineering schools.
Choten runs intensive holiday courses and one-to-one tutoring in maths and physics, from lycée (Seconde to Terminale) through the scientific classes préparatoires (MPSI, PCSI, MP, PC, PSI and more). Sessions are capped at six students, taught in Paris with a deliberately premium, close-mentoring format. Its mentors are current students of Polytechnique, CentraleSupélec, Mines Paris, Ponts et Chaussées and Télécom Paris.
choten-academie.fr ↗The method comes from the classroom. I served as an oral examiner (colleur) in mathematics at Lycée Hoche, first for an MP class then a PC class (2024 to 2026), preparing students for the Mines-Ponts and Centrale competitive exams. That is Choten's edge: mentors who are both graduates of the Top 5 and active examiners of the prépa system.
An AI voice assistant for salons and clinics that handles bookings over the phone and syncs them to Google Calendar. Sold directly to small businesses, it reached 50% adoption among the ones it went out to.
An all-in-one platform for work efficiency, co-founded with Sami Bali. Pitched at the Télécom Paris Innovation Day 2025 as one of the student start-up finalists.
The federation uniting the AI societies of France's leading universities, building a national student AI community, with partners such as Anthropic, QRT and Perplexity.
Research collaboration, speaking at your event, or just an idea worth discussing. My inbox is open.