About Me

AI Research Engineer with a PhD in Mathematics and strong foundations in probability, optimization, and linear algebra. Experienced in applied research spanning generative models, adaptive neural architectures, and dynamic networks. Comfortable reading, re-implementing, and stress-testing research papers, and translating theory into working code on real data.

Research Experience

TORUS AI — AI Scientist

Dec 2023 – May 2026

Research Theme: Dynamic & Parsimonious Neural Networks

Led company-directed R&D on topology-adaptive and expressivity-aware neural architectures.
Implemented selected approaches in PyTorch based on recent research on parsimonious and dynamic networks.
Applied models to real-world time-series data (GNSS, predictive maintenance).
Analyzed convergence behavior, architectural sensitivity, and feasibility for industrial use.

Applied Research in Generative Models

Implemented and evaluated LLMs, VAEs, GANs, UNet-based diffusion models, and transformers.
Applied parameter-efficient fine-tuning techniques (LoRA, PEFT) on LLMs and diffusion Models (SDXL, SANA).
Built complex image-generation workflows using ComfyUI for experimental evaluation (civitai.com).
Designed a speech-to-structured-data pipeline (Whisper + LLaMA 3B LoRA), generating synthetic supervision with Megatron and evaluating structured JSON extraction reliability for automated document generation.

Research Practice

Conducted structured literature reviews and presented ~8 research papers bi-weekly internally.
Regularly re-implemented methods from recent publications to validate assumptions and limits.

Selected Papers Studied

Growing Tiny Networks: Spotting Expressivity Bottlenecks and Fixing Them Optimally
Application of the Topological Gradient to Parsimonious Neural Networks
Parsimonious Neural Networks Learn Interpretable Physical Laws

Skills

Foundations: Probability, Optimization, Linear Algebra, Spectral Methods (SVD)
Research Areas: Dynamic & Parsimonious Neural Networks, Architecture Adaptation, Generative Models, Time-series Modeling
Models & Methods: Transformers, VAEs, GANs, Diffusion (UNet), LoRA / PEFT
Implementation: Python, PyTorch, Experimental ML Pipelines
Languages: English (Advanced), French (Intermediate)

Education, Publications & Honors

PhD in Mathematics — Paul Sabatier University

2020 – 2023

Thesis: Métriques hermitiennes spéciales sur les variétés complexes compactes lisses
(Involved heavy use of non-linear PDEs and Differential Geometry.)
E. Soheil, Properties of Critical Points of the Dinew-Popovici Energy Functional. Journal of Complex Manifolds, vol (9) 202.
(With D. Popovici), Functionals for the Study of LCK Metrics on Compact Complex Manifolds. Bull. des Sci. Math., 188 (2023).
Eiffel Excellence Scholarship
Bronze Medal at National Math Competition