Mickaël Vitry

**Context**: Bedrock Streaming (RTL/Bertelsmann group) operates the M6+, RTL+, Videoland video streaming platforms - 50M+ users.**Mission**: Design and implement the new data mesh analytical platform within a team of 5 (flat organization).**Challenge**: The centralized data team was a bottleneck: almost zero new data products in 2024, time-to-market of several months, integration of new clients blocked.
**What I implemented**:

• **Architecture**: Migration from a centralized data lake to a decentralized data mesh. Streaming architecture design: DynamoDB → Kinesis → Firehose → S3 → Databricks Spark Declarative Pipeline
• **Infrastructure as Code**: 7 reusable Terraform Modules (AWS: IAM, S3, KMS, Kinesis, Firehose, Lambda + Databricks: Unity Catalog, pipelines, jobs, alerts). Semantic versioning.
• **GDPR & Governance**: GDPR by design implementation (PII tagging bi-weekly, right to be forgotten within 30 days, audit trail), lineage tracking (DataHub + Unity Catalog), automatic cataloging. Member of the data governance committee.
• **Enablement**: Support for 4 domain teams (User, Payment, Catalog, Audience, Content) through workshops, pair programming, ADRs, GitHub unbooks, dedicated Slack per domain.
• **Assisted AI**: Automated doc agent (Claude integrated VSCode), planning complex migrations, PR and IaC review.

**Measurable Results**:**Stack**: AWS (Kinesis, Firehose, Lambda, S3, KMS, Glue) • Databricks (Unity Catalog, Spark Declarative Pipeline) • Terraform • Python • DataHub • GitHub Actions

**Active Pedagogy**: 10% theory / 90% practice, project-based approach focused on "learning by doing"**Taught Subjects**: ETL/ELT (Airflow, Docker), data modeling (UML, normal forms), data warehousing, big data (PySpark, distributed architecture), cloud (Azure Synapse Analytics, Data Factory, Databricks)**Personalized Support**: Individual support adapted to levels, mentoring on real projects, job preparation
**Approach**: Encourage learners to find the solution themselves rather than giving them the answer directly. Develop autonomy and problem-solving skills.**Technologies**: Python, SQL, Airflow, Docker, Azure, PySpark

**Context**: Nielsen | TVTY measures the impact of TV campaigns on the web traffic of 100+ international advertisers (Carglass, Dyson, Verisure).**Main Challenge**: Anticipating the deprecation of third-party cookies by browsers, which threatened the core of the collection model. Urgent need to redesign the tracking system while maintaining service continuity for 100+ clients.**My Mission**:

• **Critical Modernization (6 months)**: Complete redesign of the JavaScript collection tag to anticipate the end of third-party cookies. Specifications, JS development, impact study, progressive blue-green rollout with coexistence of old/new solution, client-by-client validation.
• **Extensibility**: Design and development of custom Airflow connectors for Google Analytics (GA3, GA4) and other sources. Automatic enrichment via multiple APIs.
• **Distributed Infrastructure**: Maintenance of an AWS EMR Hadoop cluster managing 100+ advertisers. Multi-tenant architecture with an Airflow DAG per advertiser, each DAG with a branch per campaign + an always-on branch for reference data.
• **IaC & DevOps**: Infrastructure as code management (boto3, Ansible, Docker), test/prod environment deployment, CI/CD with GitHub.
• **Cross-functional Support**: Data technical referent. Ad hoc analyses with customer service (EDA, Jupyter/Zeppelin visualization), production debugging, query optimization.

**Achievements**:**Stack**: Python, JavaScript, Spark, Airflow, AWS EMR, MariaDB, GitLab, Docker