Nine years of Python, PHP, Node and Angular for product teams, AI startups and business-critical applications. Today: multi-platform architectures, self-hosted RAG assistants, and the codebases nobody else wants to inherit.
I took my first steps in web development in Dakar, on a mobile payment platform that now runs in 27 countries. A passion for code and for well-built architecture took me travelling. From Clermont-Ferrand to Montpellier, I have built software that works, and helped product teams get more value out of what they ship. Today I train agents, and I ship faster than I ever have.
A complete business platform for a Martinique non-profit fighting food waste: collection route planning, geolocated field entry, certified weighing at the depot, FIFO/FEFO stock with expiry alerts, then dispatching. Angular back-office, NestJS API, native iOS and Android apps, and a RAG assistant to query the data in plain language.
Migrated a 10,000-line legacy Python app to Flask on my own, plus a CI/CD regression detection pipeline for the model. 400,000+ images used, bulk image rendering under 100ms.
Business application maintained by 13 developers and 4 POs. Migration of a 19-year-old codebase to PHP 8 and MySQL 8, with optimized schema changes.
Platform deployed across 27 countries. Built the client integration sandbox single-handedly: 4,000+ companies, 60,000+ touch points.
Generating a working API in about two minutes from a UML class diagram. The API Platform + Visual Paradigm + Venom chain, and where the line falls between CRUD services and custom ones.
A Visual Studio Code extension that turns a Visual Paradigm class-diagram export into compliant API Platform entities. Five design requirements, and the code that follows from them.
What if a single engine exposed CRUD for every entity? Java introspection to generate APIs at runtime instead of rewriting them — demonstrated on Google App Engine, over a dataset on begging in Senegal.
How to split a message between luminance and chrominance in a color JPEG image. Research at LIRMM with Marc Chaumont: at around 97% in luminance, embedding gains roughly 1% in security against the steganalyst.