Mehdi Khachichi

Contributing to the modernization of AXA’s Next Best Offer (NBO) engine within the IA/GenAI Factory by combining predictive modeling, business, and MLOps best practices to deliver scalable and value-driven AI solutions. Designed and deployed scoring models to predict customer appétence and churn risk, enabling personalized marketing campaigns and stronger customer retention. Increased targeting efficiency up to ×3 by recommending the most relevant product at the optimal time, reducing irrelevant communications. Built end-to-end ML pipelines for more than 5 million customers, from data ingestion and feature engineering to model training and evaluation on Databricks (PySpark). Applied MLOps best practices — version control (Git, Azure DevOps), reproducible experiments (MLflow), CI/CD automation, dependency management (Poetry), and testing — while integrating model explainability with SHAP to ensure interpretability and uphold high software quality standards. Benchmarked multiple approaches: Gradient Boosting (XGBoost, LightGBM, CatBoost), AutoML, and Deep Learning architectures to maximize predictive performance.

Developed and deployed an end-to-end MLOps pipeline for image denoising in the AgroConform project, enhancing agricultural imagery quality through a custom deep learning autoencoder. Built a modular and maintainable architecture using Object-Oriented Programming (OOP) and Design Patterns to ensure clean, scalable, and reusable ML components. Designed a robust data ingestion pipeline with advanced preprocessing for large agricultural datasets, optimized model performance using MSE loss, and automated experiment tracking with MLflow. Deployed the trained model as Dockerized REST APIs with FastAPI on Google Cloud Platform (GCP), ensuring scalability, reliability, and smooth integration. Applied MLOps best practices — CI/CD automation (GitHub Actions), dependency management (Poetry), version control (Git), logging, and monitoring for reproducibility and quality assurance. Followed software engineering standards: clean code, testing, type hints, and documentation to ensure maintainable, production-ready pipelines.

Successfully led IT projects from inception to completion by ensuring they aligned with client objectives and were delivered on time.

Developed a leave management application using Streamlit, integrating visualizations with Matplotlib and Seaborn alongside a MySQL database. Documented key features and provided a user guide to ensure ease of use and streamlined maintenance.