Carpooling has become an essential part of urban life. This system is highly helpful as it promotes sustainability by saving fuel energy and reduces the pollution caused by huge number of vehicles. Though there are multiple methods already devised out to handle carpooling, those systems lack in providing a clear solution. Previously researched approaches that tackled the issue used traditional Reinforcement Learning algorithms (RL) to simulate the complex environment where the shared cars operate, by employing the IDQN approach. It does learn the decentralized value but are prone to instability because of simultaneous learning and exploring done by multiple agents. The proposed method focusses on Multi-Agent Reinforcement Learning in QMIX architecture and it proves to achieve decentralized execution along with centralized learning. QMIX utilizes a mixing network that calculates joint action values as monotonic collection of each agent values. These joint action values for each agent are ensure to be non-negative because of the mixing network which guarantees consistency for different grid sizes. The proposed model supports varying count of passengers and drivers in every case proving consistency in performance in random scenarios.