Cheick Fofana

• Redesign of the engine data architecture (F1 & LMDh): Migrated from MATLAB to a modern microservices-oriented stack, reducing licensing costs and data access time. This enabled engineers to quickly monitor the evolution of critical parameters (engine wear, fuel consumption, temperature, fluid density) while ensuring robustness and observability. (Python, PySpark, Docker, Kafka, FastAPI, Azure DevOps, Grafana) • Azure Data Lake / Databricks Integration: Led the migration from on-premises to Azure, integrating over 43 TB of engine data (Parquet) and synchronizing SQL databases; implemented ingestion and transformation pipelines in Databricks following a medallion architecture, then consolidated multi-system data into a graph-oriented database to break down silos and deliver a unified view enabling more consistent and coherent analytics (Azure Data Factory, Databricks, PySpark, ADLS Gen2). • Real-time data processing optimization: Contributed to the redesign of the high-frequency telemetry (100–1000 Hz) streaming system used for on-track engine monitoring, improving service resilience and reducing data latency for engineers. The pipelines handle data ingestion, business computations, and Parquet storage. (Python, PySpark, RabbitMQ, Docker, Azure Devops) • Electrical oscillation detection via Machine Learning: Implemented an anomaly detection system on the electrical component (MGU-H) of F1 engines using telemetry data, as part of investigations into performance losses. (Databricks, Python, Pytorch, Numpy, Streamlit) • Engine test tracking tool development: Designed an application enabling system engineers to centralize and record all test bench activities, improving traceability and coordination across teams. Tech stack : Databricks, Python (Pyspark, FastAPI, Numpy, Pytorch, Streamlit), Kafka, SQL Server, Redis, Docker, Azure Devops, CI/CD, RabbitMQ, Promtail, Loki, Grafana, CosmosDB, Azure Blob Storage

• User dropout detection using Machine Learning (LSTM): Designed and deployed a predictive model on the Moodle platform based on user activity logs, leading to a 37% reduction in dropout cases after six months in production. • Skills matching (BERT fine-tuning): Developed and deployed NLP models for automatic extraction and alignment of skills (based on the SFIA framework) from profile descriptions, significantly improving the accuracy of online skills assessments. • Data architecture and ETL pipelines setup: Built the data architecture and ETL pipelines to enable daily collection of job offers, providing structured data for real-time market analysis and machine learning model training.

• Implementation of a Machine Learning–based recommendation system: Developed and industrialized a model leveraging user histories and learning paths (using Collaborative Filtering and Matrix Factorization), which improved recommendation relevance and achieved a 72% Hit Ratio after five months in production.

Realization of a state of the art around the problem of using optimization to solve Multi-Label classification problems.