Data Scientist, Quantitative Risk (apprenticeship) · Sept 2025 – Oct 2026 · Paris

This project was carried out within Crédit Agricole Personal Finance & Mobility. Data, code and detailed results are confidential: this page only describes the approach.

1- Overview

During my final-year apprenticeship in the Quantitative Risk team, I developed a machine learning model to estimate the probability of default (PD) of consumer credit customers, and took part in the migration of large-scale databases from SAS to Snowflake.

  • Role: Data Scientist apprentice
  • Period: September 2025 – October 2026
  • Stack: Python (pandas, scikit-learn, gradient boosting), Parquet, SQL, Snowflake, SAS
  • Scale: 30-million-row mixed-type dataset

2- Context & Motivation

The probability of default is a central input of consumer credit: it drives credit decisions, risk monitoring and capital requirements. The goal was to build a model that is both accurate and explainable, on a dataset large enough to raise real engineering constraints.

3- Data Preparation

  • Worked on a 30-million-row dataset mixing numerical and categorical variables.
  • Managed storage and loading with Parquet to keep memory use and processing time under control.
  • Discretized continuous variables into risk classes, the standard practice in credit scoring for robustness and interpretability.

[TO COMPLETE: number of variables, target definition (e.g. 90 days past due), observation window]

4- Modeling

  • Logistic regression as the reference model: interpretable and standard in credit risk.
  • Gradient boosting to capture non-linear effects and interactions, compared against the baseline.

[TO COMPLETE: evaluation metric used (Gini / AUC), validation method, what you can share about the comparison]

5- Data Engineering: SAS to Snowflake

As part of a data pipeline automation initiative, I migrated large-scale databases from SAS to Snowflake: converting SQL schemas and data types.

6- Takeaways

This apprenticeship gave me hands-on experience of machine learning in a regulated industrial setting: working at scale and balancing performance against explainability.

[TO COMPLETE: one concrete result you are allowed to mention]