Custom app, website, AI & digital solutions for businesses

Software Engineer · peexWeb

2022 – 2026 · Lyon, FR

Full ownership of the digital department at peexWeb alongside a colleague, covering the entire product lifecycle: from requirement analysis to production deployment, including team management (including an offshore Indian development team).

Data & ERP · Unified Back-office (Ecigplanète / Smoke Market)

Developed a custom unified business back-office from scratch using Next.js 15 / React 19 / PostgreSQL to replace a legacy 10-year-old ecosystem. This platform centralizes the management of two distinct brands: Ecigplanète (B2C) and Smoke Market (B2B).

• Key Features: 90 business modules and 450+ API routes, including real-time analytics (15s polling), revenue/margin calculators, AI-powered support system, sales management with prospecting maps, automated restocking, invoice verification, competitor tracking, and a full project management suite (Sprints, Kanban, Tickets).

• Infrastructure: Granular permission system and Big Data dimensions for aggregating and processing high volumes of business data. Internal logistics mobile apps (orders, stock, scanning, restocking) in React Native.

Generative AI & RAG · Back-office assistant and customer chatbot

Built a RAG system wired directly to live business data: a conversational assistant embedded in the back-office lets teams query orders, customers, products and statistics in natural language, scoped to each employee’s permissions. On the customer side, an AI chatbot understands the visitor’s request and recommends the right products from the catalogue. Customer support also gets AI ticket analysis (summary, customer request, suggested replies).

In-house search engine · Typesense

Built an in-house Typesense search engine for both shops, replacing a SaaS solution rented at roughly €24k a year. Multi-shop index: B2C prices indexed with tax, B2B negotiated prices computed on the fly. Gradual rollout driven by a server-side kill switch with a beta list, with no interruption for customers.

Price comparison and second-hand marketplace

France’s first platform combining a vaping price comparison site with a peer-to-peer marketplace: three-party transactions (buyer, seller, platform) with funds held in escrow until delivery. Laravel back end, Next.js front end. Press coverage in 2025. Price collection is written in Python: scraping tens of thousands of products, normalising product titles, then matching the same product across sites with a system that combines fine-tuned embeddings (Sentence Transformers), a RandomForest classifier and similarity rules.

E-commerce Overhaul · PrestaShop 1.6 → PrestaShop 9 Migration

Led the complete overhaul of two historical e-commerce sites. Unified the back-office architecture while maintaining two distinct storefronts. Migration to PrestaShop 9 (Symfony), development of custom modules, bespoke Admin APIs, and Docker configuration. Coordinated and directed a supporting team of Indian developers.

Laboratory ERP

Development of a comprehensive ERP for a manufacturing laboratory. The architecture covers recipe management, raw material stocks, and blending processes.

• Business Intelligence: Profitability analysis based on lab formulas, procurement costs, and raw material cost prices.

• UX/Role Management: Dedicated dashboards for diverse profiles (Logistics, Lab, Sales, Admin, SuperAdmin).

Quality, deployment and DevOps

Five-stage GitLab CI pipeline: static analysis (PHPStan, PHP-CS-Fixer, ESLint, Stylelint), blocking in-house convention checks, integration tests against a containerised MariaDB, blocking coverage thresholds (100% on the API, 85% on the business module). Automated deployment on a self-hosted runner: sync, autoloaders, cache purge, idempotent database migrations. Cloudflare in front of the sites: DNS, CDN, WAF and Turnstile against bots.

Monitoring of every server with Prometheus, Grafana, Loki and Alertmanager, with alerts sent to Telegram; custom metric exporters and alert rules for the search engine, each alert triggered once in acceptance testing before going live.

Cross-functional Role: Technical architecture, DevOps (Docker, servers, GitLab CI, Cloudflare), SEO, Code Quality, AI Integration, and direct stakeholder management.

Full-Stack Developer / Tech Consultant · Freelance

2024 – present

Freelance activity alongside my studies and my job. Complete management of client projects: from taking the brief to deployment in production.

EIRA Environnement

Corporate site for an environmental services company. Headless architecture Next.js 15 + CMS for client autonomy. Dynamic pages (services, careers, team), advanced SEO.

Darkdog (Garm Tattoo)

Interactive portfolio and booking platform for a tattoo artist in Lyon. Next.js 16, Prisma, PostgreSQL, 3D rendering (Three.js). 5-step booking wizard, gallery with likes, blog with SEO analyzer, admin analytics dashboard.

Eshisha-store

E-commerce site for e-shisha Prestashop, complete theme redesign, SEO optimization.

Co-founder & technical lead · CENTIO & CO-SPORT

2024 – present · Lyon, FR

Co-founder of two projects run in parallel with my studies and my work-study position.

A web and mobile platform for finding a sport partner, built single-handedly from A to Z, with over 850 users. Training log, 3D anatomical model of muscle overuse, real-time posture analysis with on-device vision including rep counting and voice coaching, 3D maps of routes and free-flight sites with weather data.

Next.js web app, Flutter app on iOS and Android, CI/CD and infrastructure run end to end.

A recruitment and scheduling SaaS for private security companies. CNAPS-verified agents, prefectoral authorisation and professional licence checked before the first application, agent-to-assignment matching based on certifications, availability and licence validity, agreements signed online with the DPAE hiring declaration generated on the spot, and no commission on shifts.

Next.js web app, Flutter app, infrastructure administered end to end (OVH, Docker, Cloudflare WAF, monitoring and alerting). Team of four, sole technical profile.

Infrastructure & real-time cockpit

I run all of my servers on my own: three Ubuntu machines at OVH and AWS, hosting four projects in production, including a client site. Nginx as a reverse proxy in front of Docker containers, daily off-site backups monitored by heartbeat, tested restores, alerts and a daily report on Telegram.

To see everything at a glance, I built a real-time cockpit. On each server, a Python agent (standard library only, systemd service) pushes a sample every 5 seconds (CPU, RAM, network, HTTP requests) and a full state every minute (containers, backups, certificates, fail2ban). A Cloudflare Worker and a Durable Object receive these samples, keep 24 hours of history in SQLite and stream them over WebSocket, on desktop and on mobile. The cockpit is hosted outside the monitored servers: if one of them goes down, it turns red instead of disappearing with it.

A 3D graph of the architecture where every particle is a real HTTP request, an nginx feed anonymised at the source (truncated IPs, URL parameters stripped), Sentry errors and HTTP probes every 30 seconds. No port opened on the servers, HMAC-signed session, one token per server, and the whole thing fits in Cloudflare's free plan.

Along the way, a few diagnoses: 61 GB of Docker build cache filling up a disk, a backup silently broken by a permissions issue (fixed, then monitored so it cannot happen again), and a Docker daemon holding on to more than 3 GB of memory after repeated builds.

Full-Stack Developer · Novius

2021 · Lyon, FR

Bug resolution, development of websites using Laravel and VueJS, creation of back-office, API integration, git usage.

Master of Science (MSc) · Information Systems Architect · Epitech

2020 – 2026 · Lyon, FR

Work-study programme. Engineering Master’s focused on Artificial Intelligence, Data and DevOps. Designed and delivered end-to-end projects: AI and language-processing systems, real-time data analytics platforms, and cloud infrastructure deployment.

• Travel Order Resolver : an NLP pipeline that turns a spoken request into an SNCF itinerary. Whisper transcription, departure and destination extraction by fine-tuning CamemBERT and spaCy, comparative model benchmark (F1, speed, confusion matrices), Streamlit interface.

• KubeQuest : a Kubernetes cluster on AWS (4 nodes): Helm deployments, Prometheus, Grafana and Loki monitoring, OIDC authentication, OPA Gatekeeper security policies, Ansible provisioning.

• Taxi Driver : an agent trained by reinforcement learning (Q-learning) on the Gym Taxi-v3 environment, hyperparameter tuning and comparison against a random baseline.

• CryptoDash : a real-time crypto analytics platform: multi-topic Kafka streaming, Puppeteer scraping, Zod validation, Next.js charts.