Nowadays, micro-mobility sharing systems have become extremely popular. Such systems consist in fleets of electric vehicles that are deployed in cities, and available for use by citizens to move in a more ecological and flexible way. Unfortunately, one of the main issues related to such technologies is their intrinsic large-scale load imbalance: Providers cannot have continuous control over the fleet, as customers can pick up and drop off electric vehicles wherever they like. We present ESB-DQN, a multi-agent system based on deep reinforcement learning (RL) that offers suggestions for picking or returning e-scooters to balance fleet usage and sharing as much as possible.