Raphael P.

I contributed to the development and structuring of an AI startup incubated at Berkeley SkyDeck (UC Berkeley accelerator). I developed document processing pipelines (RAG, GraphRAG, OCR, etc.) for their models, and contributed to the deployment of frontend features and the improvement of the Backend architecture.

Design and development of a Python learning platform for students in preparatory classes and "Grandes Écoles". Development of the web application, implementation of an automated grading system, creation of personalized learning paths, and continuous improvement of the user experience. Management of the entire development cycle, from architecture to deployment.

Research project conducted at EURECOM focusing on the analysis of language model hallucinations in RAG and GraphRAG architectures. Design of several Retrieval-Augmented Generation pipelines, creation of a dataset of 547 questions-answers, and comparative evaluation of different knowledge retrieval strategies (Vector Search, Knowledge Graphs, Cross-Encoder Reranking). The project combines academic research, experimentation, and development of evaluation tools to improve the reliability of generative AI systems.