At Institut Pprime – CNRS / University of Poitiers, I built a modular Python pipeline for a scientific dataset and fine-tuned Peak-CNN, improving its F1-score from 0.073 to 0.785. I also developed and evaluated a multi-stage 3D reconstruction workflow, documenting where detection, matching, and triangulation failed.
I built a FastAPI RAG teaching assistant with PDF ingestion, vector retrieval, and source-grounded responses, and evaluated it on 45 questions. My projects also include an agentic pentesting prototype and a medical chatbot pipeline using Qwen with LoRA.
