Droplet microfluidics has resulted in the development of functionality-based devices such as droplet synchronization, sorting, sequencing, etc., which form an essential component in the advancement of lab-on-chip technologies. The devices can passively route the droplets, i.e., without any external control. However, the application of microfluidic devices to more complex objectives requires the amalgamation of multiple objectives into a single design. One such objective is to design a single unified device to process high-throughput droplets for a given functionality. One of the most crucial functionalities for commercial microfluidic applications is droplet sorting. We present and design a droplet sorting computer with the help of a simple combined design framework. The framework is built using the Multi-Agent Reinforcement-Learning (MARL) method. The problem is highly complex; however, the desired network design is identified with the help of suitable design parameters and MARL. All the drops are routed towards one of the outlets through simple hydrodynamic interactions using a single network design with suitable inter-droplet spacing. The framework resulted in functionality-driven devices with optimized network topology and minimized inter-droplet spacing. The devices have sufficient robustness and stability for a wide range of operating parameters. The inherent property of the optimized network and spacing design enables scalability to a large number of drops with zero to minimal human intervention and viability for commercial applications. We envision that the unified device would significantly advance passive mechanisms for droplet routing across different functionalities and facilitate faster microfluidics adoption to biology and life sciences.