Mark Dakroub

Data Cloud Engineer - AMADEUS - Nice, France

(2024-10)

Migrated C++ based ETL pipelines into Azure Databricks platform reducing processing time by 40%. Deployed UAT and PDT environment for Azure Databricks enabling stable pre-production validation.
Deployed recovery pipelines or main BI projects from Scala to PySpark, improving data availability.
Developed an Azure monitoring dashboard to track PNR ingestion and new data availability, improving ETL observability by reducing manual efforts by 70%.
Delivered production-grade cost assessment for Qlik Sense Vs Power BI evaluating total cost ownership and integration feasibility within Amadeus's internal architecture.
Deployed secured Azure network infrastructure, subnets and production-ready VMs hosting Power BI Gateway to enable enterprise-grade data exchange between internal system and Power BI Service.

Data & AI Engineer - Greater Paris University Hospitals - APHP - Paris, France

(2024-03 - 2024-10)

Conducted literature review on preprocessing techniques and 3D image reconstruction for fetal brain MRI. Developed neural network models for fetal brain segmentation and detection.
Designed a data pipeline to extract 3,856 records, applied data cleaning, transformation via python. Designed a Datawarehouse via MySQL Database. Ensuring data integrity for patient's data.
Deployed and containerized medical imaging toolchains using docker and docker-compose (MRtrix3, SVRT, NiftyMIC, MONAI), enabling reproducible research environments.

AI Engineer R&D Department - Centre National de la recherche - Lyon, France

(2023-03 - 2023-10)

Built a artificial neural network for anomaly detection, analytics and ECG cardiac signal classification.
Monitored and evaluated ML models using performance metrics including F1-score, Precision, Recall, and ROC-AUC, improving predictive accuracy by 35% through iterative hyperparameter tuning and feature engineering.
Built a Datawarehouse Medallion architecture, transforming ingested data, to support downstream reporting and analytics, reducing processing time by 33%.

À propos