ShironBeskiwin

Artibox, a portfolio company of the investment fund Naxicap, builds software for the construction industry: Devibox runs franchise networks from quote to accounting, Artibox connects their calls for tender with tradespeople. The two chain together, and I work with that chain in mind.

On Devibox, run gave me a complete knowledge of the application, on which the structural projects were built: the full rebuild of the application, interface and APIs rewritten cleanly, from concept to code, driven by an ageing CRM whose ergonomics weighed on the productivity of an entire network; then internationalisation to four new countries, with their invoicing rules and currencies.

On Artibox, the stake is tradespeople's activity on the platform: multi-plan subscriptions, badges and rewards, configurable incentive banners, e-mail and notification campaigns built on shared segments. Every active tradesperson raises the odds that a tender finds a taker.

Artificial intelligence transformed our working methods over these two years. I was proactive in watching and adopting it: shared rules and skills, onboarding the team on Cursor and Claude, and an evolution of the feature development process, with measurable gains in delivery speed, test coverage and autonomy.

One Prev was building a first-level health-prevention mobile app. Five dimensions, health, mental well-being including at work, nutrition, sleep and sport, whose data is correlated to surface risk analyses. To spare the user any data entry, it was designed to draw on the phone's health data and the Apple Watch, an integration then under study.

As the sole developer, I translated the roadmap and the needs set by the CEO and CPO into technical specifications, and carried the design end to end, from architecture to UI. Six months separated the first line of code from production: the product was live well before it was complete.

Involved in strategy, I learned first-hand what an early-stage project demands: deciding what can ship within the deadline, deciding what not to build, and keeping an architecture able to grow without having been built for everything.

Air Liquide Santé cares for patients at home, notably for sleep apnea and diabetes. Within the exploratory-projects team, I worked on mapping that market from public data: starting from a few known players, rebuilding company networks step by step by analysing the links between companies and their executives, then weighting each company with the PageRank algorithm. The collection, until then ten hours a week by hand, was automated in Python and Selenium and brought down to twenty minutes a day.

Cleaning the false positives was a job in itself, because a rebuilt network is only worth the reliability of its links. The result: 322 companies attached at 92% precision, and a reading of the sector's movements, acquisitions and territorial coverage, for the strategic analysis teams.

I then crossed this network with the public Transparence Santé database, which records ties between companies and healthcare professionals, and built an ML score to estimate the strength of relationships with partner practitioners. All of it reads in interactive Power BI dashboards, built for four teams of the division.