Critical production with zero downtime
Carefully prepared go-lives, systematic rollback plans, cross-DC DRP. Proven service continuity in deposit and public service organizations.
I design and harden your critical platforms — from bare metal to Kubernetes clusters, from sovereign cloud to AI agents in production.
More than 20 years running critical Linux production, blending DevOps integration (Ansible, Terraform, GitLab CI), virtualization and sovereign clouds (VMware, OpenStack/NUBO, Kubernetes/Onyxia), and now AI in production. Cross-functional technical reference at the French Ministry of Finance, Radio France, the National Library of France, INPI.
They have trusted me
20+yrs
of Linux production experience
500+VMs
operated in critical environments
8missions
major · public, media, finance, energy
3datacenters
migrated with no service downtime
Profile
French citizen — based in Paris. Technical reference on critical environments: bridge between production and development, delivery automation, security and large-scale operations (hundreds of VMs, monitoring, storage).
Partnerships
In addition to my own assignments, I regularly collaborate with partners to address broader needs (cross-functional teams, GPU infrastructures, AI, etc.).
Value proposition
A rare profile: the technical depth of a Linux systems expert, combined with the rigor of a DevOps integrator who knows how to ship.
Carefully prepared go-lives, systematic rollback plans, cross-DC DRP. Proven service continuity in deposit and public service organizations.
From rack mounting and Fibre Channel SAN to Kubernetes clusters and S3 / Ceph storage. End-to-end visibility few profiles can claim.
Ansible, Terraform, Jenkins, GitLab CI, Helm. I turn fragile chains into reliable deliveries — and I document what I do.
I support project managers, unblock teams on Linux/AD, virtualization and ANSSI security, and pass on my know-how.
LLMs integrated into my tooling: Cursor, n8n + Claude agents, local Ollama / Mistral inference, RAG. Concretely: AI agents in production at a small business, legal RAG delivered at a hackathon, and this portfolio as a data-driven build.
AI · agents · inference
AI is not a fad for me: ever since accessible LLMs appeared, I've made them both a daily working companion and an experimentation field. I use them to code, design and reason — and I build agents that run in production, including local inference to preserve data sovereignty.
AI-augmented IDE
my main development environment, with built-in agents and automation
Main LLM assistant
design, analysis, documentation, refactoring, code review
Research & reasoning
tech monitoring, benchmarks, multi-source synthesis
Contextual auto-completion
occasional complement on GitHub-hosted projects
Agent orchestrator
document sorting and processing pipelines, triggers and API calls
Recognition & extraction
invoices, accounting documents, structuring of non-standard documents
Local LLM inference
running on workstation / server, for data sovereignty
Open-weight LLM
models served locally via Ollama, first deployment in a small business
ASR — speech to text
Feb 2025 hackathon: voice input for the legal chatbot
Contextualized RAG
answers grounded in the French Legal Code (sources from git.tricoteuses.fr)
REST API
Python layer linking ASR, RAG engine and rendering
Concrete projects
Career
From the French Ministry of Finance to Radio France, via the National Library of France and INPI: a dense career in critical environments, on a wide range of topics.
Ministry of Finance — Bercy HUB & DGFIP
Two entities
Assignments with at least two entities of the ministry: Bercy HUB for the Nubonyxia project, then the DGFIP (taxes) for the RADAR project.
Bercy HUB — Nubonyxia project
The Nubonyxia project relies on providing the Onyxia software — born at Insee with contributions from Bercy teams. Onyxia installs on a Kubernetes cluster and deploys workloads as pods via Helm charts. The offer rides on the chosen hosting base (Bercy HUB / NUBO), hence the project name (Nubo / Onyxia).
Role: contributing to the operations of an installation already in production, plus rolling out additional services. In practice: adapting Helm charts so they can be launched from the Onyxia portal — turning packages into the Onyxia flavor to comply with the catalog model and compliance requirements.
An automation chain was already in place to move deliverables from development through to user availability. I was responsible in particular for the healthy operation of this continuous integration chain, with many topics to handle around authentication and security.
One major challenge: mapping all components and flows to make them visible and controllable — the perimeter was hard to scope while authentication raised many issues and continuous integration chains remained only mildly stable.
DGFIP — RADAR project
RADAR project goal: provide a framework that aggregates several inventory sources in order to map the versions of the structuring components of the information system (visibility into the fleet and into what is actually running in production).
Role of application DevOps integrator on this perimeter: Ansible, Nexus, Jenkins, GitLab tool chain, with services including CMDBUILD and Apache NiFi, complemented by Tomcat, PostgreSQL, on a Linux base on NUBO / OpenStack. Deployment relies in particular on Jenkins and Ansible.
On arrival, the platform showed numerous malfunctions. My role was mostly to stabilise what had already been delivered — including writing and maintaining Ansible and Terraform code for creating virtual machines on OpenStack — then to drive the version upgrade of all RADAR components — including the move of CMDBUILD to its latest version (alignment with the integration chain and dependencies).
In the absence of a dedicated JavaScript developer, I also took part in maintaining and evolving the RADAR UI, built with the Ext JS framework.
Enedis, Fayat IT
Enedis — Jan 2025
Designed a test bench for real-time software: verified the chaining between Linux and the real-time application.
Fayat IT — Jan 2024
Worked on the migration of a piece of software from the Linux environment to Windows.
RYC — business support
Since 2005, supporting RYC, a business assistance structure (advice on administration, pre-accounting notably with Sage Coala).
IT and office support: Windows workstations on the client side; Linux servers for infrastructure, with folder sharing via Samba — the Coala software running on the Windows workstations.
Accounting flow automation
For about five years, strong rise in automation: automatic extraction of bank statements and semi-automatic integration into the accounting software up to the posting of accounting entries.
Very recently: automatic extraction of invoices and attachments to generate accounting entries, also relying on bank reconciliation.
AI, agents and orchestration (very recent)
The arrival of AI opened up new possibilities: designing AI agents, using n8n for document sorting and processing, calling APIs — notably Claude (Anthropic) — for the recognition and processing of documents, extending the existing automation.
First experience with local inference of two models via Ollama (running on workstation / server), based on Mistral LLMs.
DGC — training centre (same group as EPSI)
Support and student fleet
During studies at EPSI, student job at DGC, a training centre in the same group as the school.
IT support and fleet management of the workstations dedicated to students: Windows NT 4, then Windows 2000. Created images with Ghost (Symantec / Norton Ghost) and multicast deployment to re-image machines regularly; Windows profile management (many incidents to handle); permissions and access for student accounts; antivirus installation. Also handled training software / e-learning deployed on site.
Also the chance to set up Linux on recycled hardware to provide extra workstations serving site usage.
Tech watch
Selection of projects I find compelling to understand where the systems, virtualization and browser ecosystem is heading. Good illustration of today's frontiers between Linux, low-level and WebAssembly.
The Linux kernel booting directly in the browser via WebAssembly: BusyBox + musl, Xterm.js terminal. A fascinating proof of concept for anyone interested in scheduling, system primitives and the limits of the modern JS sandbox (no MMU, task suspension emulated via Web Workers, etc.).
Discover the project Emulation · WebAssemblyDOS / Windows 95-98 emulator in the browser, based on DOSBox-X compiled to WebAssembly via Emscripten. Browser-side persistent hard disk, ISO/IMG/CD handling, gamepad support — a very convincing demo of what Wasm with exceptions and asyncify enables today.
Discover the projectWebAssembly
Pick a tab: both demos are served from this site (DOS and Linux Wasm) with browser-friendly headers. The downloads further down remain optional.
Repo download, Emscripten build and publication under /wasm-lab/: detailed in wasm-lab/BUILD.md. Script: npm run wasm:fetch.
DOS Wasm: /wasm-lab/deploy/ — After npm run wasm:fetch, the emulator files are under wasm-lab/deploy/ (index.html at the root). Push as-is to the server, under the same tree.
Projects
In parallel with my assignments:
In 2026, leading the creation of a web portal for managing artwork collections, with AI assistance.
In February 2025, took part in a hackathon dedicated to artificial intelligence: our team built a voice chatbot able to extract and quickly serve information from the French Legal Code. Goal: simplify access to law through speech recognition and synthesis.
Legal sources — the reference texts came from git.tricoteuses.fr, providing a complete legislative base.
Technologies — Whisper (voice → text), LightRAG (contextualized RAG on GitHub), shell scripts to clean and format datasets, Python & FastAPI to expose web services, dataset preparation on a development machine (Mac mini, 24 GB RAM).
Infrastructure — LLM deployment on the Kubernetes GPU cluster of SPESYS Services, with GPU access for performance suitable to the short hackathon time frame.
My role — dataset preparation: extraction, structuring and normalisation of legislative texts; creation of Python/FastAPI primitive APIs linking the speech recognition layer, the RAG module and the rendering.
Collaboration — two-day hackathon with exchanges with the DINUM, the Bercy HUB team around the Onyxia (Nubonyxia) project, and other public-sector actors (Ministry of the Economy, Ministry of Justice, etc.).
Acknowledgements — Stéphane Baisse and the SPESYS teams (Thomas Williot, Gérald Moreno) for infra access and support.
Innovation — accessibility (questions out loud), GPU-powered speed, social impact for professionals and the general public.
Since 2023, developing a software portal for data scientists, focused on data science and AI.
In 2022, volunteer participation in the creation of an art institute in Échirolles, as IT expert (advice and rollout of the digital base).
In 2020, built websites with WordPress and Elementor for small entrepreneurs.
Le Signe — artwork collection management software: Groovy (backend), JavaScript and React (frontend). Tested and developed around 2019 as part of collection inventory at Le Signe, French national centre for graphic design in Chaumont.
In 2013, designed an automatic MCQ correction solution built on the open source Auto Multiple Choice (AMC) tool: generation of unique questionnaires per exam session (questions and answers in a different order from one copy to another), unique barcode per printed paper copy, then scanning, automatic answer recognition (OMR) and automatic grading.
Skills
Synthesis aligned with my career: Linux/Unix operations in public service, finance, media and energy; virtualization (VMware to Proxmox), private OpenStack/NUBO clouds, Kubernetes & Helm (Onyxia, public-sector compliance), HPC Slurm/Apptainer; automation with Ansible, Terraform, GitLab, Salt; ANSSI hardening, MCS and Cyberwatch; AI agents and document chains in production.
Linux Red Hat/CentOS, Debian/Ubuntu, SUSE in production (level 3); Active Directory / LDAP / SSSD integration (Radio France, Naarea); PXE boot / Preseed, LTSP for thin clients (UCAD); BIND DNS, iptables firewall; kernel compilation and trimming (recycled fleet). Unix AIX, Solaris, HP-UX (INPI, Sungard GP3 migrations). Windows and Samba in mixed contexts (small businesses, museum).
Centreon, Grafana, Prometheus; Nagios → Centreon → Prometheus journey (INPI, Naarea). Graylog, Elastic Stack for logs and correlation. JMX metrics (Tomcat/Java). Tech-functional operations dashboards.
HP 3PAR SAN, iSCSI, Fibre Channel, NFS; Ceph, S3 / MinIO object storage; MySQL Galera + ProxySQL. VMware Site Recovery Manager DRP / BCP, cross-DC replication and datacenter migrations (INPI). Backups: Bacula, BackupPC, NetBackup, Veeam.
VMware vSphere, oVirt, KVM, Proxmox, Hyper-V; OpenStack (NUBO, ministry). Kubernetes & Helm (Onyxia / Nubonyxia, "Onyxia-flavored" charts, catalog CI). Docker; Apptainer for container-style workloads; first clusters via Rancher / RancherOS (INPI).
MySQL / MariaDB, PostgreSQL, Oracle (operations), MongoDB, MaxDB. Tomcat / Java stacks, Apache NiFi, CMDBuild (RADAR DGFIP). PHP, Node, Heurist (SHS at BnF) integrations. Ext JS (business UI).
Ansible (Tower), Terraform (OpenStack VMs), Puppet, SaltStack; Git, Jenkins, GitLab CI, Bercy/BnF/INPI release chains; Rundeck → Ansible Tower (Radio France). Dollar Universe (scheduling). Bamboo / SVN (Sungard era).
Bash/shell, Python, JavaScript/React, Go, Ext JS; Django, PHP, VBA/AutoIt. Recent projects: FastAPI, RAG, Whisper. In production: n8n agents, Claude API, local Ollama / Mistral inference (small business), operations scripts and Selenium (prod checks).
Slurm, InfiniBand, Apptainer (MPI, scientific workloads), Lenovo platforms; protected network zones, dedicated LDAP (Naarea). Linux master images hardened to ANSSI guides, MCS. Cyberwatch (INPI rollout, Radio France advice).
Education
2004
Master's degree (Bac+5) — Information Systems Expert
Typical path: French baccalaureate — BTS — EPSI program (LIS, DGC…) — Master's / IS expertise.
Engineering internships: ISTA, STMI, METO X SILICIO, LFB, LIS, DGC (1999–2005). To be completed in index.html or in this JSON, then npm run build.
Documents
Resumes and presentations as printable HTML (PDF), plus the raw data (JSON) and a structured text profile for ATS and recruiter AI tools.
Synthetic and dense resume that fits on a single A4 page. Best for classic applications, IT staffing recruiters and first contacts.
Open the raw fileSpacious multi-page format, polished and readable presentation. Ideal for IT recruiters and headhunters who want detail.
Open the raw fileMy profile from a human angle: stance, soft skills, values, commitments and narrated career path. Free of jargon, for a first interview.
Open the raw fileExhaustive technical stack, deployed architectures, signature diagnostics and frameworks. For those who want to assess expertise on the substance.
Open the raw fileFaithful copy of data/site.json: same source as the showcase and generated resumes. Ideal for importing the profile into another system or for automated processing.
Open the raw file Pretty view (colored / readable)All-text version, headings and lists, metadata in header: designed to be read by matching tools and LLMs integrated into ATS.
Open the raw file Pretty view (colored / readable)Contact
For a discussion about an assignment or a need for Linux / integration expertise.
age 7
First computer: PC XT with 8088 processor at 8 MHz, 512 KB RAM, two 5.25-inch floppy drives (360 KB), 84-key keyboard, CGA monitor, MS-DOS 2.21; also discovered the Logo language; first at home, our first database-style software: a family genealogy program.
age 8
Basics of GW-BASIC; MS-DOS and creation of batch files (.bat).
age 10
First time disassembling a computer, then full reinstall on my own after wiping data: floppy formatting, hard disk formatting, configuration of CONFIG.SYS and AUTOEXEC.BAT.
age 11
Introduction to DTP (desktop publishing), as part of classes in Paris.
age 12
Learning Turbo Pascal and first video game development; first steps in C.
age 13
Created a computer club at middle school; first discovery of the Internet at university.
age 14
Second computer: 486 DX at 40 MHz, 4 MB RAM, 3.5-inch floppy drive, 250 MB hard disk, VGA monitor; running MS-DOS 6.2 with Windows 3.1.
age 16
First Linux installations on my personal machines; first kernel compilations; sharing the internet connection via ipchains.
age 17
First LAN parties.