Axel Michalewicz

Context:

Within CNEPE (EDF), as part of the implementation of the Group Data Platform (PDG), aimed at standardizing the exploitation of data (industrial, IoT, visualization). The objective was to ensure the reliability and industrialization of pipelines for acceptance and production.

Team:

Technical referent for 2 Data Analysts, with a lead role in Data Engineering and orchestration.

My role:

I:

- supported the team in upskilling on Data Engineering best practices

- conducted a complete audit of Airflow pipelines (structure, performance, maintainability)

- corrected and stabilized approximately 15 existing DAGs

- implemented a standardized methodology for DAG development

- separated business logic through a dedicated Python library

- optimized S3 -> Apache Iceberg / Postgres (Bronze / Silver / Gold) flows

- introduced dbt to ensure transformation reliability

- implemented CI/CD (GitLab CI) with unit and integration tests (pytest)

- implemented native Airflow monitoring and an email alerting system

Results:

- Complete stabilization of pipelines → reliable operation

- Development of a DAG in max 1 day thanks to standardization

- Significantly improved maintainability (logic decoupled from DAGs)

- Accelerated incident detection and resolution

- Technical foundation ready for acceptance and production

- Team upskilling on Airflow and ETL practices

Languages: Python, SQL

Tools - DB: PostgreSQL, Apache Iceberg, dbt, boto3, pytest

Infrastructure: AWS S3, Airflow, GitLab, GitLabCI

Context: The mission was to create the database schema for a startup developing a SaaS simulation tool for health insurance. The first clients began using the platform: they entered their data, launched simulations, and the results had to be stored in the database for quick access and later analysis.

Team: As the sole Data Engineer, I worked with a full-stack developer who handled the rest of the platform, and the two founders who formulated the needs and priorities.

My role:

I:

- designed the database schema from scratch and all the necessary PostgreSQL tables,

- implemented the ORM with SQLAlchemy and the Alembic migration tool to ensure schema evolution reliability,

- configured the development and execution environment using Python, SQL, PostgreSQL, SQLAlchemy, Alembic, and Docker.

📈 Results:

- Implementation of persistent storage for simulation results: previously, each result had to be recalculated manually, causing several minutes of waiting per simulation.

- Instant access to existing results for users, greatly improving their productivity.

- Correction of query and typing issues, eliminating several sources of errors in calculations.

- Tens of simulations now historized, allowing for reliable tracking and better data organization.

- Guaranteed traceability: the migration tool allowed clear tracking of schema evolutions, where no visibility existed before.

Languages: Python, SQL, bash

Tools - DB: PostgreSQL, SQLALchemy, Alembic

Infrastructure: Git, Docker

Context: Airseas is developing a kite sail prototype to reduce ship fuel consumption. During the testing phase, the Data team's challenge was to provide reliable, complete, and quickly usable flight data to aeronautical engineers to accelerate prototype development.

Team: I worked within a small Data team consisting of a Product Owner and myself, in daily contact with an Ops systems engineer for infrastructure and about thirty aeronautical engineers, the end-users of the data.

My role:

As a Data Engineer, I:

- set up and industrialized flight data ingestion pipelines via Airflow, reducing manual ingestion efforts and accelerating data availability a few hours after a test,

- automated post-processing using a dedicated calculation library,

- improved real-time streaming by extending Telegraf connections to new machines.

📈 Results:

- Ingestion time reduced by 80% (from several days to a few hours).

- Aeronautical engineers' analysis accelerated by 50%: flights ending at night, data was ready by morning for post-processing.

- Real-time streaming with < 100 ms latency, improving monitoring and decision-making during tests.

- Over 100 flights processed automatically, ensuring a reliable and continuous chain.

- 80% reduction in ingestion failures due to reduced manual interventions.

Languages: Python, SQL, bash

Tools - DB: PostgreSQL, DuckDB, Parquet, Grafana, MQTT, Telegraf, TimescaleDB, JupyterHub

Infrastructure: Docker, Airflow, GitLab, GitLabCI, Rundeck