Thomas L.

• End-to-end management and development of a client data migration automation platform (multi-client/multi-project environment) to accelerate the transition from legacy IT systems to a standardized target.

• Design of a self-service workflow allowing business/consulting teams to configure imports and mappings directly within the tool, reducing technical back-and-forth and manual processing.

• Full-stack development: React front-end (configuration, mapping, steering) + Python back-end (orchestration, execution, control).

• Orchestration via Apache Airflow with dynamic DAG generation per client/project/entity, to industrialize execution and improve run traceability.

• Addition of an LLM-assisted component to accelerate the production of transformation SQL (generation + manual editing).

• Automation of internal processes for Product Owners, particularly to reduce repetitive tasks in monitoring, formalization, and project coordination.

• Implementation of assisted release note generation from product evolutions, to accelerate communication to stakeholders and ensure delivery traceability.

Synthetic data generation (tabular) for banking use cases — model, API & app.

Last assignment in R&D at Skaizen Group (remote from Tokyo): scoping, applied research, and delivery of a complete stack to produce realistic, usable, and secure datasets.

On the research side, I conducted a targeted state-of-the-art review (tabular GANs, diffusion models, graph ML) with critical analysis and concrete recommendations for transactional data (sender/recipient pairs, temporality, amounts, currencies). Objective: to balance statistical quality, diversity, and confidentiality, then translate these choices into product design.

On the engineering side, I:

• Designed a custom generation model and its API;

• Developed a multipage web application to control generation, visualize the structure of synthetic datasets, and evaluate their quality using adapted tabular/graph metrics;

• Integrated encryption/pseudonymization of sensitive variables for controlled sharing.

On the deployment side, I industrialized on AWS (API on EC2, exchanges via S3) to orchestrate front↔back flows and pave the way for rapid POCs in client environments.

Impact. The solution allows to:

• Test models in near-real-world situations without exposing personal data;

• Accelerate R&D (varied and traceable synthetic datasets);

• De-risk data access and facilitate product-business-security collaboration. I also co-authored a conference paper related to this work.

Large-scale Data Engineering — from raw signals to analytical datasets.

Mission: transform massive, heterogeneous, and difficult-to-use datasets into reliable, compact, and queryable data for all teams.

• Volume & performance. Normalization of raw formats to a unified schema + compression/structuring strategies for a 10x reduction in disk footprint, without data loss.

Design of Python/Dask ETL pipelines to process data volume in a distributed manner and accelerate computation times.

• Signal valorization. Extraction of interpretable features (events, aggregates, time windows) from vehicle data to feed R&D and product teams. Close collaboration with specialists to prioritize valuable information.

• Industrialization & sharing. Implementation of strategic partitioning and naming conventions for more efficient queries and batch jobs. "Ready-to-use" datasets exposed to internal teams without degrading source systems.

• Quality & reliability. Data testing, consistency checks, usage documentation, and schema contracts to ensure long-term viability and reusability at scale.

• Agile framework (SAFe). Active participation in ceremonies, multidisciplinary synchronization, and continuous delivery focused on quality, scalability, and robustness.