Bonjour Maxence,
comprendre le francais c'est pas de problème, mais pour les choses techniques c'est plus simple pour moi parler l'anglais.
I am a software developer and data scientist from northern Germany, currently working in a research group about ML and AI in the context of university education.
As far as I can see, AWS Textract's purpose is to process scanned documents like classic OCR technology does. Reviewing the given PDFs, most of them are not scanned, but directly contain text data that can simply be extracted without any (pixelwise) processing.
But let's discuss this in a private chat. I'd be interested in evaluating Textract as well, as one of my customers' projects deals with document mangement.
Thanks,
Jörg