At Berger-Levrault’s AI Lab, I implemented an adaptive memory and user-preference personalisation system for an LLM agent platform for maintenance management. I modelled 10 preference dimensions and built a prompt evaluation pipeline with 500 test cases, integrated into GitLab CI.
For my master’s project, I fine-tuned sentence encoders for narrative similarity, reaching an F1 score of 0.981. I’ve also studied database performance at IRIT and developed a Node.js backend and responsive web interfaces at MetalMaker 3D.

