We have analyzed the seismic series that preceded La Palma 2021 eruption using a purely automatic methodology based on Deep Learning. A new catalog has been retrieved, firstly 5797 absolute locations were obtained using NLLoc algorithm and considering a 3D velocity model of the island. At a further stage, these results were improved computing relative locations with HypoDD code. Our final catalog has doubled the number of events of the original manual catalog of the Instituto Geográfico Nacional. We observe new detailed features of the seismicity migration throughout the crust. We can clearly differentiate two phases within the sequence: the first one from September 11th to 16th and the second one from September 16th until the eruption onset. This differentiation can be described in terms of independent seismic parameters: inter-event time dispersion coefficient, b-value spatio-temporal distribution, hypocentral migration, magnitude distribution). Our results are in agreement with previous geodetic studies. Finally, we have developed a conceptual model describing the magmatic unrest and how the magma made its way to the surface prior to this eruption.