jordan Ndetcho

𝘾𝙡𝙞𝙚𝙣𝙩: ALLIANZ 𝘾𝙤𝙣𝙩𝙚𝙭𝙩: Allianz’s accounting division required stable, auditable, and performant ETL workflows to handle financial data pipelines related to personal insurance operations. The objective was to maintain and improve existing workflows, as well as develop new data ingestion and transformation processes using Python and PySpark, with orchestration via Luigi. 𝘼𝙘𝙩𝙞𝙤𝙣𝙨:   • Maintained and optimized existing data pipelines, improving readability, performance, and robustness.   • Developed new PySpark jobs and Python scripts for data ingestion & processing from sources like PostgreSQL and MongoDB.   • Implemented pre-checks and validation mechanisms to detect errors early in the pipeline.   • Collaborated within a small agile team (2 DE and 1 PM), ensuring continuous delivery and minimal downtime. 𝘾𝙝𝙖𝙡𝙡𝙚𝙣𝙜𝙚𝙨: The biggest challenge occurred during monthly closing cycles, where all pipelines had to run correctly and pass validation checks under time pressure. Failures required urgent root-cause analysis and corrections, as missed deadlines could delay or compromise critical reporting and financial operations. 𝙍𝙚𝙨𝙪𝙡𝙩𝙨:   • Successfully stabilized end-of-month data processing, reducing the risk of reporting delays.   • Improved pipeline transparency and reliability, easing troubleshooting and team handoffs.

𝘾𝙡𝙞𝙚𝙣𝙩: EDF 𝘾𝙤𝙣𝙩𝙚𝙭𝙩: EDF’s internal data platforms were running large-scale batch workflows to process sales cycle data. The mission was to develop, maintain, and optimize ETL pipelines and Spark jobs that transform and aggregate data stored in HBase, write results to Hive, and export it via Sqoop into Oracle systems — where it would then feed PowerBI dashboards and internal applications used daily by business units. 𝘼𝙘𝙩𝙞𝙤𝙣𝙨:   • Maintained and refactored existing Spark and Hive workflows to improve reliability and execution speed.   • Developed new data pipelines in response to evolving business needs and working in agile environment using SAFe methodology   • Collaborated in a data squad composed of 3 DE, 4 BI developers, and 1 PM, ensuring seamless data availability for BI tools.   • Used tools such as Spark, Hive, HBase, Sqoop, and Oracle to handle full pipeline lifecycle (from raw data processing to dashboard consumption). 𝘾𝙝𝙖𝙡𝙡𝙚𝙣𝙜𝙚𝙨:   • A key challenge was ensuring night batch jobs (the system handled terabytes of data) were optimized to run efficiently and avoid failures, especially given they ran in parallel with other processes. 𝙍𝙚𝙨𝙪𝙡𝙩𝙨:   • Improved system robustness and reliability, ensuring critical PowerBI dashboards and applications received timely, accurate data.   • Saved time for BI teams and end-users through better data freshness and reduced incidents.

𝘾𝙡𝙞𝙚𝙣𝙩: SUEZ 𝘾𝙤𝙣𝙩𝙚𝙭𝙩: As part of its infrastructure modernization effort, Suez launched a project to migrate 26,000 workstations from Windows 7 to Windows 10. However, the project was stalled due to a lack of visibility into software installations and no clear framework for prioritizing sites or machines for deployment. 𝘼𝙘𝙩𝙞𝙤𝙣𝙨: > Designed and developed a Python CLI tool that:   • Analyzed SCCM and AD data to map applications to machines, solving the visibility issue.   • Computed machine eligibility per site and user.   • Recommended migration scheduling based on site readiness and deployment staff availability.   • Automatically triggered email notifications to users to schedule their upgrade once approved. > Developed a PowerBI dashboard that allowed the central deployment team to track migration progress site-by-site. > Worked closely with the planning team, deployment team, and internal stakeholders to ensure seamless tool integration and user onboarding. 𝘾𝙝𝙖𝙡𝙡𝙚𝙣𝙜𝙚𝙨:   • Cross-referencing data between machines, users, and applications without duplicates was quite tedious.   • Data freshness and accuracy was crucious to maintain a good vision of deployment progress 𝙍𝙚𝙨𝙪𝙡𝙩𝙨:   • €1M+ cost savings by avoiding manual machine checks.   • Enabled the project to restart and successfully migrated all 26,000 machines on time and within budget.   • The CLI tool became the central operational asset for planning the migration, and the dashboard provided transparency and confidence for stakeholders.

𝘾𝙡𝙞𝙚𝙣𝙩: THALES 𝘾𝙤𝙣𝙩𝙚𝙭𝙩: Following Microsoft’s end-of-support announcement for Windows 7, Thales launched a digital transformation project to migrate 40,000 computers across multiple sites to Windows 10. The company needed a smart, scalable tool to identify eligible machines for migration and plan deployments according to logistical, security, and organizational constraints. 𝘼𝙘𝙩𝙞𝙤𝙣𝙨:   • Designed and developed NOEL (Nouvel Outil d'Éligibilité et de Lotissement), a Python-based CLI tool that identifies eligible machines and groups them into optimized deployment batches.   • Implemented an optimization algorithm to maximize eligible migrations while respecting constraints (site readiness, deployer availability, machine priority, etc.).   • Built an interactive PowerBI dashboard for real-time monitoring of deployment progress, segmented by site.   • Wrote documentation for deployment planners and supported the project lead in client meetings (COPIL) and coordination with infrastructure and local IT teams.   • Used Python, PowerBI, Git, Docker, and internal APIs to ensure robustness and integration. 𝘾𝙝𝙖𝙡𝙡𝙚𝙣𝙜𝙚𝙨:   • The main challenges involved reconciling heterogeneous datasets from SCCM and Active Directory, ensuring data accuracy, and designing a system that could account for operational constraints such as: site readiness, deployment staff availability, device priority and security restrictions   • Non-standard machines not listed in SCCM => need data reconciliation 𝙍𝙚𝙨𝙪𝙡𝙩𝙨: The tool was instrumental in the successful migration of all 40,000 machines within the tight project timeline. It allowed Thales to:   • Accelerate planning and decision-making  • Ensure high data accuracy and accountability  • & Track deployment progress

𝘾𝙡𝙞𝙚𝙣𝙩: EDF 𝘾𝙤𝙣𝙩𝙚𝙭𝙩: EDF, as part of an innovation program, sought to enhance its prospection efficiency by leveraging predictive analytics. The goal was to develop a machine learning algorithm capable of estimating the likelihood that a commercial quote would be accepted, based on historical data. This would support marketing teams in prioritizing opportunities and optimizing pricing strategies. 𝘼𝙘𝙩𝙞𝙤𝙣𝙨:   • Led the end-to-end development of the predictive algorithm, including model selection, training, and evaluation.   • Used logistic regression and random forest classifiers in R and Python to model price elasticity and quote acceptance probabilities.   • Built a custom RShiny web application to operationalize the model, allowing marketers to input key parameters of a quote and receive a success probability and score (defined as: expected gain × probability).   • Integrated the solution into an Agile Scrum team, version-controlled via Git and deployed using Azure infrastructure.   • Delivered full documentation to facilitate future iterations. 𝘾𝙝𝙖𝙡𝙡𝙚𝙣𝙜𝙚𝙨: The main challenge was the quality and structure of historical data, particularly dealing with seasonal variations that complicated feature engineering and scope selection. Data sparsity and inconsistent labeling also created friction during the model training phase. 𝙍𝙚𝙨𝙪𝙡𝙩𝙨: Delivered a successful proof of concept demonstrating predictive analytics capabilities for EDF's marketing team, enabling faster, data-informed decisions. This has led to a follow-up project focused on enhancing data quality based on insights gained from the POC implementation.