9 years building systems that hold.

The spark — 2016

During my studies, I wanted to understand how a server could hold hundreds of concurrent players without breaking. Nobody around me could explain it, so I built my own architecture. C#/C++, binary serialization, hand-rolled TCP/UDP sockets. The first version ran on a €30/month dedicated server.

From 2016 to 2019, it stayed a personal project on the side of my classes. In 2019, I registered it as a sole proprietorship: the server was now sustaining 30,000 accounts and 250 concurrent connections at all times. CI/CD, proactive monitoring, autoscaling — everything arrived as problems showed up, not from a checklist. I led a 15-person team around that project, with no management degree: just the obligation to ship every week in front of real users.

The technology acquired by Rockstar — 2023

I left the project in 2021. Two years later, the framework was merged with a competitor and then acquired by Rockstar Games. I had been off it for a long time — but the architecture had held and kept scaling without me. That's the real validation: a system designed well enough to survive its author's departure.

The corporate years — 2017 to 2025

In parallel and then full-time, I built a classic corporate experience: apprenticeship at MDP Qualité on geospatial SaaS, then TECHFORM (2D/3D CPQ configurators, profiling of invoicing pipelines). Then EXFO in Quebec, on fiber/5G field apps — Flutter, offline-first, low-level Bluetooth / Wi-Fi protocols.

At Gecko Alliance, I dropped one level deeper in the stack: C / FreeRTOS firmware under extreme memory constraints, MQTT → AWS IoT Core telemetry pipelines for 150,000+ systems in production. That mission gave me the full-stack view from embedded firmware all the way up to the cloud dashboard.

Today — independence & native GenAI in .NET

Since 2025, I work as an independent on projects where the architecture conditions the product. OneRP: a lock-free engine for 2,048 concurrent connections, zero allocation on the hot paths. Poisson Engine: 100,000 entities simulated at 60 FPS in a browser via WebGPU. SaleCast: an Event-Driven SaaS platform with native ML in C#. PromptVault: a reversible PII anonymization pipeline for LLM use (ChatGPT, Claude, Gemini, Mistral). aiSelector: a 5-provider LLM gateway with a double healthcheck and model drift tracking.

On those products, GenAI is integrated natively in .NET— Semantic Kernel, Microsoft.Extensions.AI (MEAI), Agent Framework, the official OpenAI and Anthropic SDKs, ML.NET for prediction and explainability (TreeSHAP), ONNX Runtime, pgvector for semantic search. It's an uncommon angle on the market: most .NET architects outsource to Python, most AI engineers don't touch .NET. I keep both worlds in parallel.

My CodersRank rankings — Top 0.9% C# globally, Top 0.7% TypeScript, Top 5 France Entity Framework Core are not trophies: they are the trace of nine years writing serious code on real problems, not tutorials.

What I'm looking for

• Permanent role — Lead Engineer, .NET Software Architect, AI Solutions Architect or AI Engineer in a team that ships to production. Indicative: €65 — 95k gross / year depending on role, equity, architecture scope. Remote-first, hybrid Occitanie or Île-de-France OK.
• Freelance / consulting — architecture missions, native GenAI integration in .NET, performance audits, IoT / firmware. Day rate €600 — 900 ex. VAT depending on context, duration and criticality. Listed on Malt.
• What I'm not looking for — pure ML research / fine-tuning, frozen legacy code with no room to evolve, missions where I would be the only person able to understand the code. A good architecture must be readable by its successors.

How I work

I start from the real context: production code, allocation profile, measured contention. No off-the-shelf prescriptions. When I step in, the goal is always double: solve the immediate problem, and transfer the skill to the team — so it does not come back once I'm gone.

I turn down missions where I would be the only one able to understand what runs. A good architecture must be readable by its successors.

Off-code

Based in Lunel (Occitanie). Available remote for France and Europe, on-site in Occitanie and Île-de-France. I read a lot about distributed performance, lock-free architectures, and compute shaders — it's my way to keep learning when the day-to-day work no longer raises hard questions.