Freelancer profile translated to English.

Description

Senior Data Scientist with over 20 years of experience, I help companies transform their complex data into reliable, high-performing, and business-actionable models.

My core expertise: tabular machine learning, risk/fraud scoring, model optimization, leakage detection, feature engineering, ML pipeline auditing, interpretability, and lightweight industrialization in Python.

I am also a Kaggle Grandmaster, ranked among the world's top machine learning competitors. This experience has taught me to extract the maximum from challenging datasets, to challenge metrics, to detect methodological flaws, and to rapidly build robust solutions.

I am particularly effective in:

auditing or improving existing models;

benchmarking ML / AutoML solutions;

fraud, risk, propensity, churn scoring, or anomaly detection;

model performance optimization;

explainability and methodological validation;

AI / data science technical scoping;

supporting data teams on complex topics.

I have worked in demanding environments: telecom, banking, insurance, fraud, marketing, AutoML platforms, and industrial data projects. My goal is simple: to deliver useful, measurable, and understandable results quickly, without unnecessary over-complication.

Preferred missions: short audits, model optimization, AI scoping, ML prototyping, business scoring, Kaggle-like challenges, expert support.

Preferred format: full remote, short or medium-term missions, high added value.

Industry field of expertise

Languages

French
Native or bilingual
English
Conversational

Workplace preferences

Remote only

Primarily works remotely

AdvanThink (ISoft)
Senior Data Scientist
SOFTWARE PUBLISHING
December 2022 - Today (3 years and 8 months)
Pays de la Loire, France
- Optimization of anti-fraud models
- AI Watch
- Development of AI features in the tool
- Mentoring junior data scientists
- Demos / Pre-sales
- AI services for our clients / prospects
- Participation in Kaggle competitions
Risk / Fraud Scoring Data Modeling
Orange
Senior Data Scientist
TELECOMMUNICATIONS
April 2016 - July 2019 (3 years and 3 months)
84100 Orange, France
integration of new methodologies to manage the entire upstream/downstream machine learning pipeline regardless of the data source, as well as the integration of new vertical business modules,
- Enrich product APIs to make it modular,
- Promote the product's reputation and user communities (in conjunction with marketing operations),
- Expand the user community, establish educational resources. Orange France 15 years 2 months Data Scientist April 2016 - July 2019 (3 years 4 months) Arcueil
- Customer dissatisfaction scoring (Customer Experience Management project) => implementation of an R Shiny application presenting the top explanatory variables per customer.
- Support for Orange Bank scoring projects (pre-attribution scores / wealth scores), DMP (online cookie scoring), fraud detection.
- Beta Tester for the automatic modeling solution Predicsis / Khiops, including multi-table modeling (star schema). Solution available standalone, Cloud (Service as a Solution), soon compatible with Hadoop/Spark.
- Development of an API automating modeling processes with various software (Kxen, Khiops, Statmining, R Xgboost and H2o, Python Scikit learn, Vowpal Wabbit) and algorithms (about thirty algorithms including: GBM, RF, DP, GLM, NB, SVM, KNN, ET) => addition of Keras, Regularized Greedy Forest to the study
- Development of about fifty Amadea operators including: N stratified folds, Symbolic Preprocessing: One Hot Encoding, symbolic to factors, symbolic to count, symbolic to target rate, Numeric Preprocessing: Box Cox transformations, log, t-SNE, numeric to percentiles, Feature Selection (near 0 variance, duplicate columns, high missing rate, xgboost features importance), Replace NA by score, Feature engineering: N-way interactions, counts, metadatas, clustering, dimension reduction, Blending / Stacking. Niche research
Amadea Python Technical Management
Orange SA - Orange Digital Ventures
Data Scientist
TECH
January 2014 - April 2016 (2 years and 3 months)
Montrouge, France
- Within the Digital BI project, involvement in 2 POCs: (Erwan Le Nagard), characterization of the Orange community on social networks (Facebook and Twitter), (Nicolas Gilot) campaigns on customers opted-in from their Facebook data, program for matching Facebook and Orange customers. 2015 Project: establishing a partnership with Facebook?
- Exploration of new Hadoop-compatible tools for accessing the Hub France (2014: Actian, 2015 project: Talend, Ab Initio, Dataiku) to identify tools that allow us to work in both Hadoop and Teradata. Within the Explorers program, testing new data sources available in Hub France (2014: mobile data, 2015 project: CEM).
- Within the UTE project, testing the contribution of TV usage data: design of the test plan for incorporating this data into propensity scores, 180 generalization scores for all households (individual and/or multi-person), setting up the interface with the Smart Data Factory, highlighting the relevance of propensity scores versus thematic attraction scores/markings, deployment of propensity scores including data from TV logs starting in November 2014. Early 2015, further enrichment with VOD, SVOD, and TVOD usage data. Project objectives:
- Creation of over 500 additional scores to improve customer knowledge by generalizing known (or usable because opted-in) information to all households (individual and/or multi-person), available upon request
Technical Management Data Science Machine Learning Algorithms