Freelancer profile translated to English.

Description

Specialized in DeepTech development, I'm the one called when "off-the-shelf" solutions are no longer sufficient.

My job? Running complex algorithms where they shouldn't be running.

I am a Systems Architect with an obsession for native performance and data sovereignty. I like to know how things work underneath (and often, rebuild them).

The Lab(Recent R&D Projects)

I spend my nights experimenting. Here's what's coming out of the garage:

GPU Optimization (Velvet): I coded a custom CUDA optimizer. Basically, it's AdamW on steroids (-13.8% loss at startup, still).

Cognitive Architecture (NeuroVA): A high-performance RAG system in Rust. The idea was simple: an AI that has memory without needing to rent a data center for €10k.

Simulation & 3D: I make cubes move with Deep Reinforcement Learning in Unreal Engine. It's a start, but they learn fast.

Audio & DSP: A native Float32 audio engine, because sound is serious business.

💡 Why work with me?

Because I'm a radical self-taught learner.

I didn't learn code in a classroom 5 years ago. I learn it by hitting walls.

I deliver code that compiles, runs fast, and is documented.

My preferred stack:

Languages:Rust(memory safety),C++(for raw performance),Python(for prototyping),CUDA(for the pain).

Tools: **PyTorch**, **OptiX**, **Unreal Engine 5**.

Disclaimer:

I don't sleep much, I consume a suspicious amount of coffee, and I tend to consider a compiler bug a personal insult.

If you're looking for someone to build a WordPress site, I'm clearly not your guy. If you have a somewhat scary technical project, we'll get along well.

Available forR&D / Proto / MVPmissions.

Industry field of expertise

Languages

French
Native or bilingual
English
Native or bilingual

Workplace preferences

Remote only

Primarily works remotely

The Vesper House
Lead Developer - Professional Audio Editor (Contract)
November 2025 - Today (7 months)
I led the development of VADE-R from start to finish, a native and high-performance desktop application for non-destructive audio editing.

Key Achievements:

Software Architecture: Design of a Float32 audio processing pipeline for maximum quality, and development of a non-destructive editing engine with history (undo/redo).

Full-Stack Development: Creation of the complete application (+14k lines of code) with Electron, React, and TypeScript, ensuring robustness and performance.

UI/UX Design: Design of a complex and intuitive user interface for multi-track editing, real-time effects application, and video project management.
Typescript React Native Software Architecture Branding Node.js
The Vesper House
Founder & Software Architect
October 2025 - Today (8 months)
As the founder of **The Vesper House**, I transform complex technical challenges into elegant and high-performance software solutions.
My role is to lead the entire product lifecycle, from fundamental research and development to the design of polished user experiences.
I act as a technological partner for my projects, bringing cross-functional expertise in Artificial Intelligence, C++/Python development, and real-time 3D tool creation.
My mission is to build custom tools that are not only powerful but also intuitive and enjoyable to use.
CUDA Python C++ Typescript Machine learning
The Vesper House
AI Architect & Research Engineer
August 2025 - Today (10 months)
I led the design and implementation of **VelvetAI**, a next-generation Artificial Intelligence architecture.
This R&D mandate focused on performance optimization and the creation of innovative models.

Key Achievements:

GPU Optimization:
Development of a custom C++/CUDA optimizer ("**VelvetOptimizer**"), demonstrating over 10% performance improvement compared to standard solutions like AdamW.

LLM Architecture: Design of "VesperLM", an advanced language model integrating cutting-edge techniques (FlyLoRA, Mixture-of-Experts) for increased parameter efficiency.

Training Pipeline: Setup of a complete system, from dataset curation to training and inference, withauto-scalingtools based on **Chinchilla laws**.
CUDA C++ Python Pytorch Data Analysis