Strong gravitational lenses have turned out to be cosmic jewels serving a lot of astrophysical and cosmological applications.There are various kinds of strong lenses : we will focus here on galaxy-galaxy systems, showing extended arcs around a central galaxy. When combined with dynamical methods that break the mass-sheet degeneracy they suffer from, these lenses constraint the dark matter fraction and the dark matter profile of the deflector galaxies. Then, the derivation of such constraints for an entire sample of lenses can tell us about the evolution of galaxies and eventually discriminate between the different theoretical scenarios. To finish, the statistics of lenses enable us to have advanced knowledge either about the galaxies number and structure, or cosmological parameters such as $\Omega_{m}$ and $\Omega_{\Lambda}$ (or even $H_{0}$ if the extended source hosts an intrisically variable object). Unfortunately, strong lenses are relatively scarce objects, so that the issue of detection remains crucial. Until there, the modelling of lensing systems was finely performed downstream from their detection, then based on spectroscopic or photometric methods. We have imagined a new fast and automated detection tool which sets modelling at the very core of the selection process : the development and optimization of this Robot-based approach, that is presented here, thus lies in a subtle balance between speed and robustness.