Benoist Légal

• Designed an end-to-end Data MVP combining recipe, seasonality and agricultural production datasets to generate personalized and sustainability-oriented recipe recommendations. • Built data ingestion, cleaning, matching and quality-control workflows to reconcile heterogeneous ingredient and agricultural data. • Designed a recommendation scoring approach combining recipe relevance, seasonality, agricultural availability and other recipe attributes. • Built and automated a Bronze / Silver / Gold data architecture using Python, Airflow, dbt, Parquet, MinIO/S3, DuckDB and FastAPI.

YUMMY – Sustainable Recipe Recommendation Platform

• Analyzed datasets using SQL and Python to investigate anomalies, assess their impact and support corrective actions across Data & AI projects. • Performed data quality, completeness and consistency checks on large volumes of media transcription and reference data, contributing to more reliable downstream processing and data delivery. • Analyzed and mapped end-to-end data flows across the data.ina.fr ecosystem, from media capture and AI-based transcription to data processing, quality controls and final restitution. • Gathered and analyzed business requirements, formalized business rules and specifications, and translated operational needs into requirements understandable by technical teams. • Worked cross-functionally with Data Engineers, Data Scientists, developers, project managers and business users, supporting projects from requirements analysis and data investigation to testing and validation. • Contributed to Data PoCs involving data qualification, APIs and multiple reference systems, assessing data consistency and supporting the functional design and validation of solutions. • Developed a Python training-data pipeline for a BERT-based named-entity disambiguation model. Analyzed false positives and introduced hard negatives to improve the quality of the training dataset and model precision. • Automated integration and regression testing using JMeter and Jenkins, reducing control time by more than 50% and improving repeatability of validation workflows. • Contributed to the industrialization and deployment of Data processing pipelines using Dagster, Jenkins CI/CD, Git, Docker and Helm.

Institut national de l’audiovisuel (INA)