Leo Chartier

Data Scientist Llm, Vlm @EDF

Ville de Paris, FR

Project: Sensitive Content Detection for Nuclear Documents (EDF)LLM Fine-Tuning with LoRA: Fine-tuned large language models (LLMs) using LoRA techniques to detect and explain sensitive content in EDF nuclear power plant documents, ensuring compliance with European nuclear export control regulations.Large-Scale Document OCR: Pre-processed approximately 2 million documents by evaluating multiple OCR solutions and ultimately selecting Azure OCR for optimal performance.Data Labeling & Clustering: Enhanced data labeling by clustering document pages using multimodal Vision-Language Models (VLMs) and computing cosine similarity based solely on the vision component of pre-existing models.Alternative VLM Approach: Explored fine-tuning the entire VLM with LoRA directly on PDF documents, although this approach was less effective compared to the LLM fine-tuning on OCR-processed texts.

Université Paris Cité

Master 2 (M2), Mathématiques et informatique

ENSAI

Diplôme d'ingénieur, Statistiques