Deep networks (DN) perform cognitive tasks related with image and text at human-level. To extract and exploit the knowledge coded within these networks we propose a framework which combines state-of-the-art technology in parallelization, storage and analysis. Our goal, to make DN models available to all data scientists.