Lead Data Scientist

Covéa Data Science Team since Oct. 21 : NLP, ML/Deep-Learning in Production with Azure + DataBricks

3.5 years at Technical Excellence Center AXA Global P&C: Improving pricing techniques & using machine learning techniques for claims handling. Also getting & leveraging external data for better pricing, user-experience & claims management.

3.5 years at French Ministry of Health : Modelling Health Sequences, handling Big Data from the Public Health Insurance (covers 100% of 67M french citizens). Also tailoring tools to gather & visualize data for decision making during CoViD-19 crisis.

Lead Data Scientist NLP

Data Science Internal Consulting Team

Paris, France

2021-Now

• Multiple NLP projects in Production Tackling Labelling, Annotation, Model & Data Monitoring including Data Drift detection, MultiLabel Topic Classification, Sentiment Analysis and Anonymization using fine-tuned RoBERTa-like Transformers with HuggingFace with Azure + Databricks.

Senior Data Scientist & (EIG)

Statistical Departement at Ministry of Health

Paris, France

2018-2021

• Modelling Health Sequences to predict diseases outcome and detect disruption in the treatment course.
• Classification with active-learning & General Public WebApp to find Health-focused statistics.
• Tools for regional health agencies : Interactive Decision Making Tool to help experts elaborate Zoning for health professionals.
• Tech Lead at CoViD-19 crisis center : develop webapps to gather critical information from hospitals - ventilators and BioLabs - supplies, tests results, screening centers location & general info. Share data to stakeholders through dashboards & make advanced statistics from full hospital (SIVIC) & screening (SIDEP) data.
• Produce open-data on hospital admissions data (PMSI) involving privacy constraints k-anonymity & hierarchical l-diversity.

Actuarial Data Scientist

AXA Global P&C

Paris, France

2014-2017

• P&C pricing innovation for both housing & car insurance through zoning, vehicle classification, severity/frequency/propensity modelling with gradient boosting techniques transferred to GLM using ML-interpretation techniques.
• Build a Claim Cost Analyzer by predicting the theoretical cost of a claim - for a given vehicle & crash - to score a car repairer given their own case-mix. PoC with AXA-Spain, deployed in Spain then adapted to Italy & France with local Data Engineering teams.
• Leveraging French Court Decision Open-Data through Natural Language Processing to better handle bodily injury cases and assess contentious risk.
• Roads own-risk assessement based on GPS telematics data.

Actuarial Thesis

AXA Belgium

Remote

2014

Handle 1M contracts pricing-database to measure ceteris paribus impact of eldering on car crash severity & frequency using econometrics. Elderly drivers own-risk assessment Final selection for SCOR prize.

Long Internships

SCOR P&C, then Exane Derivatives.

Paris, France

2012-2013

6 month pricing CAT-Bonds with MCMC techniques 6 month building a synthetic index as a dynamic basket of stocks & bonds

Deep learning

Advanced training & hands-on projects

fast.ai, datascientest, deeplearning.ai

2017-2020

deep learning training : Computer Vision and Natural Language Understanding. Mainly use transfer-learning / fine-tuning. But also train models from scratch for Health Sequences Modelling using pseudo-NLP techniques: LSTM & 🤗 transformers.

Spark & Scala

Scala Programming + Spark applications

Coursera by Martin Odersky & Heather Miller

2019

While following this MOOC, I used pySpark on a daily basis on a High Performance Computer at work to handle National Health Claims Data (SNDS).

Web Development

Introduction to HTML, CSS, Javascript & JQuery.

CodeSchool.com

2016

Training + application using MEAN-Stack : Mongo, Express, Angular & Node. Develop a fast-quote API for Housing Insurance with Express. This training has been very helpful to develop advanced Data-WebApps with R-Shiny.

Introduction to data science

Main techniques of supervised & unsupervised learning.

Coursera by Bill Howe

2014

Support Vector Machine, Gradient Boosting, Random Forests, k-means & hierarchical clustering.

ENSAE Paris - IP Paris

MSc Actuarial Science & Data Science

Paris, France

2010-2014

Learning both Data-Science + Big-Data techniques & applications to Actuarial Sciences

Data Science Teacher

Data-Science for Actuaries & Data Science Certification

Paris, France

2014-2020

• Natural Language Processing & Text-mining techniques
• Machine Learning for structured data : Gradient Boosting & Support Vector Machine
• Data Science Hands-On with R, data.table, xgboost, glmnet & liblinear…
• Build Interactive Apps with R + Shiny
• Data Science for Actuaries (DS4A): 5-days training with hands-on & 1-day hackathon to teach AXA actuaries Data-Science Techniques that can help them better solve Insurance-related problems. The Theoretical Training was given by Arthur Charpentier.