I build models, then measure how much to trust them.
Final-year engineering student in statistics and data science at ESSAI, Tunis — actuarial science track, ranked fourth in my cohort, admitted through the national competitive examination after two years of preparatory classes in mathematics and physics. Looking for a six-month final-year internship from March 2027, in France or Europe.
Python · R · SQL · Monte Carlo · stratified sampling
Field survey, international benchmarking and a five-year forward simulation, run during a research internship at Hammamet Valley Hub. The estimate comes with its interval, because a single number would have hidden how far the projection actually moves.
Python · NumPy
Matrix algebra, gradient descent and backpropagation implemented without a high-level framework, then validated on real data. Written to understand what the frameworks do, not to replace them.
LangChain · LangGraph · Selenium · Streamlit
An agentic pipeline converting scraped HTML to Markdown, then to structured CSV — the shape of the output governed by the user's prompt rather than a fixed schema. Running as a Hugging Face Space.
Python · WoE / IV · IFRS 9 · macro stress scenarios
A scoring system and a loss-given-default model built to the IFRS 9 expected-credit-loss framework: variables discretised by weight of evidence, selected by information value, provisioning measured against stressed macroeconomic paths.
Academic — code on request
R · rvest · ggplot2 · k-means · HAC
Automated collection of meteorological records for fifteen cities since 2015, then time-series analysis and classification of those cities by climatic profile. The dataset is published openly.
Hugging Face · Transformers
A language model fine-tuned to detect and correct network anomalies, benchmarked against the heuristic baselines it was meant to replace.