Aleks Dominik Nougbele

Context

Mission within Enedis's data teams aimed at modernizing data pipelines and industrializing forecasting models used for electricity distribution management.

Team

Collaboration with data teams and business teams responsible for electricity consumption forecasting.

My Role

I participate in the modernization of data pipelines and the production deployment of prediction models.

Migration of historical R analytical scripts to Python to standardize the data ecosystem

Development of data and ML processing pipelines with PySpark

Production deployment of an electricity load prediction script by commercial profile

Orchestration of pipelines and tasks with Airflow (DAGs)

Integration of external data (weather API) to enrich forecasting models

Management of execution environments and industrialization of workflows

Implementation of CI/CD pipelines for linting, testing, and deployment

Tools / Environments

Python, PySpark, Airflow, PostgreSQL, Teradata, Kubernetes, GitLab CI, Polars, SQLAlchemy, MlFlow

Results

Industrialization of prediction pipelines and improved code maintainability through migration to a standardized Python stack.

Context

Development of L'OréalGPT, an internal generative AI platform allowing group teams to use various LLM models in a secure and industrialized environment.

Team

Collaboration with AI, data, and cloud teams to develop and deploy the GenAI platform at the enterprise scale.

My Role

I participate in the development of the backend and the integration services for LLM models into the internal ecosystem.

Specifically, I am working on:

Integration and benchmarking of several LLM models (DeepSeek, Gemini, Claude)

Development of backend services enabling model usage via API

Implementation of prompt generation and management modules

Experimentation with RAG architectures using LangChain

Management of model access through GCP IAM roles

Implementation of monitoring and logging to analyze usage and performance

Tools / Environments

Python, FastAPI, LangChain, Transformers, GCP, Vertex AI, Cloud Run, Firestore, Docker

Results

The platform is currently used by over 13,000 employees of the group to leverage LLM capabilities in various business use cases.

Development of a photovoltaic savings estimator for marketing purposes.

- Development of features using Flask in an AWS Lambda deployment environment.

- Code refactoring and optimization.

Integration of PVGIS, Google Solar Insight, and Google Solar DataLayer APIs.

- Creation of interfaces using React.js.

- Integration of reCAPTCHA V3 validation on React.

- Implementation of a JWT authentication mechanism based on Google reCAPTCHA validation.

Technical Environment: React.js, reCAPTCHA v3, RTK Query, Redux, Shadcn, Docker, Flask, AWS Lambda.