As part of its mission to manage feedback from experience (REX), the STRD/GQTE service has a large documentary base made up of thousands of documents. This documentation consists of technical feedback reports (REX), in PDF form (ocerized) and a “WIKIMEDIA” type knowledge base. Large Language Models (LLMs) have recently shown their ability to assist engineers in writing technical documentation, with tools like ChatGPT. RAG (Retrieval Augmented Generation) type algorithms combine knowledge retrieval techniques specific to the business domain with the generation techniques provided by LLMs. RAG techniques make it possible to supplement the shortcomings of LLMs: reduction of hallucinations, citation of data sources, and better linking between facts extracted from different documents.