Mathieu Woroniecki
AI Architect & CTO
Interactive 3D version ↗Download as PDFVersion française
Profile
AI is at the heart of my work in two ways: I build it into the products I design, and I work with it every day. Behind that, more than ten years on the web, from front-end integration to CTO.
I build AI in. Inside the product: documents read, data extracted and checked, answers that cite their sources, agents that act within set bounds.
I work with AI. Every day: dozens of agents I orchestrate, tested and measured, to ship at the pace of a ten-person team.
Currently CTO of MARKO, an AI-native B2B SaaS for real estate.
- 1 developer + his agents = a team of 10
- 10+ years of experience
- 150,000+ users on a platform I led
- 85+ projects delivered as a freelancer
Method · One developer, the strength of a team
Where it used to take a team of ten developers and several years, I ship alone, surrounded by agents I orchestrate. AI doesn’t replace the craft: it multiplies the reach of whoever knows how to direct it. Here is how I work.
Dozens of terminals at once. I no longer code one thing at a time: dozens of workstreams run simultaneously, each in its own terminal and isolated Git worktree, with its own agent. While one writes, others test, review and document.
One person, a whole team. A lead agent breaks the goal down and launches specialized sub-agents: exploration, code, tests, review. They all work at once and report back; I put it together and make the call.
Custom skills and plugins. As soon as a task recurs, it becomes a tool: skills, commands, dedicated agents, MCP servers. On MARKO, a full security audit runs with a single command.
Test everything, measure everything. Harnesses, orchestrators, plugins, extensions: I test what comes out, compare on real projects and keep only what saves time. Some experiments became published tools.
A fleet of agents on a single product. On MARKO, large workstreams go through Claude Code’s multi-agent workflows (ultracode), which spread the work across many agents; quick decisions go to Jev (TypeSafe AI), a new kind of model that doesn’t write but decides: a typed answer with its probability, in a few hundred milliseconds. The tooling changes every month; the method stays.
Speed without losing control. Tests at every step, automated review in CI, security scanners, and a human decision before every merge. The agents propose; I approve and sign off.
The proof: MARKO
12,046commits in eight months
2.1Mlines of code
95%of commits: me and my agents
Not an estimate: MARKO’s Git history, as of September 28, 2026. The code isn’t shown, only what it took to build it.
Selected projects
MARKO · The operating system for real estate debt. An AI-native B2B SaaS for asset managers, funds, family offices and crowdfunding platforms: financing documents become data, covenants are monitored, and reporting goes out on its own. marko.fr
Archon · Investigating across thousands of documents. A local investigation platform: it ingests PDFs, images, emails, archives and disk images, reads and indexes them, then lets you search, date, connect and query everything in natural language.
NumerusX · AI agents versus the market. An experimental algorithmic trading platform on Solana: AI agents analyze, decide and manage risk together; strategies evolve through genetic algorithms, and every decision is explainable. A research sandbox, not investment advice.
ScanRift · Security scanners, an LLM, and always a human. A security assessment dashboard for authorized targets: it brings together well-established open-source scanners, rates each finding, then an LLM helps analyze them and weed out false positives. Nothing runs on its own: every action is launched and approved by hand.
NumOSINT · Open-source intelligence tools, in one place. A web platform that orchestrates several open-source OSINT tools, each in its own container, and cross-references their results into investigations and case files. Built for investigation and defense.
SafeShare · Redact a document without ever sending it. Drop in a PDF or an image: the tool detects sensitive data, applies masks and exports a flattened copy. Everything happens in the browser. An AI agent can co-edit the masks via WebMCP without ever seeing the sensitive values. mathieuworoniecki.github.io/safeshare-webmcp/?demo=1&lang=en
HUman · “What if we were the largest LLM ever observed?”. A reflection on AI told in India ink and particles: from a word inside a model to the flower, the atom, the universe and the brain. It runs entirely in the browser. areweai.dev
Other projects : codex-crew · marko-support · NumerusAI · endless · CursorAuto · Algoritmi · Webiteasy · teuteuteu.com · 3D product page
Skills
AI & Data
LLM integrationsGemini · Mistral · Claude · OpenAI / Codex · DeepSeek · Local models (Ollama)
RAG & searchHybrid vector + lexical RAG · Reranking · Embeddings · pgvector · Qdrant · ChromaDB · Meilisearch · RAG evaluation
AgentsAgentic AI · MCP / WebMCP · Multi-agent orchestration · Sub-agents · Skills & plugins · Prompt engineering
DocumentsDocument AI · OCR (Mistral, Tesseract) · Data extraction · Entities (spaCy, LangExtract)
Machine learningscikit-learn · LightGBM · Optuna · MLflow · SHAP / LIME · Genetic algorithms · Monte Carlo
Front-End & Interfaces
FrameworksReact · Next.js · Vue · Nuxt · TypeScript · JavaScript ES6+
UITailwindCSS · Radix / shadcn · TanStack Query · Zustand · Redux · Pinia · Vuetify
Animation & 3Dthree.js · Canvas · Web Audio · GSAP · D3 · Phaser
ToolingVite · Webpack · Sass · i18n (50+ languages, RTL) · Web performance · Chrome extensions
Back-End & Data
LanguagesPython · TypeScript / Node.js · PHP · Go · Bun
FrameworksFastAPI · Async SQLAlchemy · Pydantic · Express · Prisma · Symfony · Headless WordPress
Jobs & real timeCelery · Dramatiq · Redis · Socket.IO · Colyseus
DataPostgreSQL · MySQL · Supabase · InfluxDB · MinIO / S3 · SQLite
DevOps & Cloud
ContainersDocker · Docker Compose · Podman · Traefik · Nginx · Caddy
CI/CDGitHub Actions · GitLab CI · Dagger · Canary deployments · Renovate
CloudAzure · AWS · Vercel · GitHub Pages
ObservabilityPrometheus · Grafana · Loki · OpenTelemetry · Sentry
Security & Quality
SecurityOWASP · CodeQL · Semgrep · Trivy · Gitleaks · CycloneDX SBOM · EdDSA JWT · Rate limiting
Personal dataPrivacy by design · Redaction · Per-client isolation · IP address hashing
Testingpytest · Vitest · Jest · Playwright · Cypress
AccessibilityRGAA AA · WCAG
Leadership & Method
TeamTech lead · Management (up to 5 engineers) · Mentoring · Fractional CTO
Augmented developmentMulti-terminal, multi-agent · One Git worktree per agent · In-house skills and plugins · Continuous harness benchmarking
MethodAgile · Gitflow · Architecture decision records (ADR) · Industrialization · Refactoring
StrategyTechnical roadmap · Pre-sales · Consulting
Experience
02/2026 — present
MARKO · CTO
Founded MARKO, an AI-native B2B SaaS for real estate investment professionals: consolidating portfolio data, automating monitoring, producing investor reporting.
- Architecture and development of the entire AI system
- Generative AI, LLM / RAG, Agentic AI, Document AI
- One PostgreSQL database per client, observability and security in CI
02/2024 — 02/2026
LWA · Lead Developer, Vue.js / Nuxt.js Architecture
Technical strategy for luxury brands (Hermès, Chanel, Ardian); team of 3 developers.
- Adopted generative AI (LLM, RAG): +35% productivity
- MCP for the Figma-to-code workflow
- 15+ projects moved to Docker: +40% deployment frequency
- GitLab CI/CD and agile: +25% velocity
03/2022 — 03/2024
ENGIE · Tech Lead, Web Platforms
Technical lead for the Digiplace strategic intranet and the public website: 150,000+ employees, CAC 40 environment.
- 99.99% uptime
- React / headless WordPress: −40% load time
- OWASP: 100% of critical vulnerabilities fixed
- RGAA level AA accessibility compliance
08/2020 — 07/2021
Sodexo · International Tech Lead, Acquisition Platforms
Global acquisition platforms (WordPress / PHP on Azure): 25 countries, 5M+ visitors a month, Fortune 500 environment.
- Legacy code overhaul: −30% technical debt
- Core Web Vitals: +45% on average
- Azure CI/CD: 99.9% uptime
- Incidents across 25 countries: resolution time cut by 4 h
11/2018 — 05/2020
AXA · Front-End Developer, UI/UX & Conversion
At the crossroads of design, engineering and results: React, Webpack, Symfony.
- +25% conversion on landing pages
- −800 ms average load time
- Internal tools used by 10 departments
03/2018 — 03/2022
Malt · Senior Consultant, Fractional CTO
Freelance, alongside other roles: architecture, full-stack development and strategic consulting.
- 85+ projects delivered, 4.88/5 across 85 reviews
- 40+ API integrations (Salesforce, Zoho, ActiveCampaign)
- Fractional CTO for 15+ startups and SMBs
01/2016 — 02/2018
1min30 · Lead Developer & Technical Manager
Inbound marketing agency: WordPress developer, then head of the technical division.
- Team of 5 engineers, 50+ web projects a year
- Industrialized production: −30% development time
- Technical pre-sales: $250K in new contracts
09/2014 — 12/2015
Indexel · Web Developer, Front-End Integrator
Built 100+ brochure and e-commerce websites (HTML5, CSS3, JavaScript), WordPress and Drupal.
10/2012 — 08/2014
Schneider Electric · IT & Technical Support
An equipment failure analysis system across multiple sites: from the equipment to the database, from diagnosis to resolution.
Education · Languages
2014 — 2016 Master’s degree (RNCP Level 2) — Multimedia Project Management, Institut F2I
2012 — 2014 BTS SIO (Associate degree in IT), Software Development track (SLAM), ITIC Paris
2009 — 2012 Vocational Baccalaureate SEN, Computer Science, Christophe Colomb
French (native) · English (fluent, technical) · Spanish (basic)