Both ammonia (NH3) and phosphine (PH3) play a significant role in an extensive range of industrial processes, while they are harmful to human health even at very low concentration. So far, a variety of gas sensors have been developed to detect them in an industrial environment aimed to protect the health of workers at their work place. Among various types of gas sensors, chemiresistive type gas sensors have attracted considerable interest due to its characteristics, such as simple fabrication, high sensitivity, high reliability, etc. Nevertheless, there are still some limitations, such as, high power consumption resulted from high operating temperatures, and most sensors are solely dedicated to an individual gas monitoring. In this work, we present the development of highly sensitive and highly discriminative graphene-based gas sensors for gas detection and identification at room temperature. Graphene is exfoliated by a liquid phase approach and functionalized by copper phthalocyanine derivate (CuPc). Leveraging machine learning techniques, graphene-based gas sensors demonstrate an excellent gas identification performance towards NH3 and PH3 at an ultralow concentration (ppb level). This work could pave the path to design highly sensitive and smart gas sensors for a wide range of gases.