At Aiway, I engineered GenAI applications and RAG pipelines for document classification and automated field-filling in French notarial workflows. I upgraded legacy fastText baselines with domain-adapted transformers and open-source LLMs, improving F1-score by 32%.
I built PostgreSQL and pgvector pipelines for semantic search and context retrieval, and created an AI evaluation framework using Golden Datasets. I also established MLflow infrastructure for experiment tracking and reproducible evaluation.
Earlier, as a Research Intern with M-PSI Team, LIG, I developed the COCOR benchmark dataset to evaluate computer vision robustness to orientation shifts. At Spans, I developed responsive React applications and interactive visual modules, and mentored a sub-team of two frontend developers.
