I take data the whole way — from scattered sources to a clean model, a clear dashboard, and a call someone can act on. End to end, tested, and honest about what the numbers say.
End-of-studies internship in a Data & Analytics consulting team. Five client missions across health, energy, tech and territorial economics — from raw data to decision-support deliverables.
An analysis is only worth what reaches the person deciding. Here is the same care-home engagement, delivered to three audiences — one dataset, three levels of detail.
The board wants a decision and a number. The business team wants the reasoning, so they can push back on it. The technical team wants to re-run the computation and check the assumptions.
Writing all three is not a formatting exercise: each time, it means deciding what actually matters to the person reading.
Built on my own time, shipped and used daily. Click a card for the full story.
Six areas I've worked across — each grounded in projects I actually built, tested and deployed.
Messy data into clean dashboards, maps and reports that surface the story.
Clustering, regression and anomaly detection — validated, not overfit.
Scraping, cleaning and joining 5+ sources into one reliable dataset.
FastAPI backends, interactive front-ends and desktop tools.
Linux servers with systemd and cron, secrets kept safe — including a self-hosted LLM running on a free cloud instance.
A scientific method that keeps the analysis honest — hypotheses tested, biases hunted, ideas invalidated with numbers.
Python, SQL, Elasticsearch, FastAPI, Ollama…
Nothing for show. Everything actually ran.
Research, tools, dashboards and experiments — click any card for the details.
Six years, three steps. Click a step for the details.
Second year of the Data Science master's, taught across École Polytechnique, ENSAE and Télécom Paris.