FIDELIA project
A major breakthrough in language assessment thanks to AI
FIDELIA is a project led by France Éducation international (FEI), in collaboration with the CENTAL laboratory of the Catholic University of Louvain (UCLouvain) in Belgium and Coexya. It represents a major advancement in language proficiency assessment by integrating artificial intelligence to enhance the correction of written responses in the computer-based Test de Connaissance du Français (TCF).
The CENTAL laboratory at UCLouvain developed a functional prototype of the solution. Coexya then stepped in to transform it into a robust, scalable, and high-performance solution capable of meeting FEI’s operational needs.
A groundbreaking innovation in French language assessment
The FIDELIA solution aims to improve the quality and fairness of assessments for candidates while optimizing the correction management process by combining artificial intelligence with human grading.
AI plays a central role in this project by automating and enhancing the reliability of written test corrections. FIDELIA integrates into the current TCF process, which is based on two independent corrections per candidate. In case of significant discrepancies, a third human correction is conducted to validate the final result.
This hybrid model values human expertise while benefiting from the precision of AI, which has been rigorously trained on France Éducation international’s data. It ensures better standardization, objectivity, and reproducibility of assessments.
The key role of Coexya in the development of FIDELIA
In this project, Coexya played a crucial role in making the original solution more reliable and scalable. Its contributions include:
An audit of the prototype’s functionality
and the underlying models developed by the CENTAL laboratory at UCLouvain.
Optimization of the code
to enhance robustness and performance.
Structuring the application
to facilitate deployment, maintenance, and scalability.
Implementing a real-time monitoring tool
based on Apache Superset, enabling oversight of the correction process as well as the analysis and tracking of the performance of the applied models.
A solution compliant with current regulations
Before its deployment, FIDELIA underwent an impact assessment to ensure its compliance with the General Data Protection Regulation (GDPR) and the European AI regulations. A bias study, leveraging psychometric analyses aligned with the latest advancements in language assessment, validated the solution’s fairness for candidates.
This initiative is part of an ongoing innovation approach based on AI, positioning FEI at the forefront of international language assessment.
