5G standardization is going to reach its end, so research on 6G has started, driven by the scientific and industrial communities. 5G and especially 6G, are going to provide resources to enhance every aspect of human life via communication networks and computing. Among the different verticals, emergency services are one of the most important parts of making smart cities of the future a reality. In this context, firefighting is highly important for security and safety. However, firefighting requires ultra-reliable and low-latency communications since firefighters can be provided with advanced guidance during fire control with the employment of distributed sensors, robots, etc. Also, the use of machine learning algorithms is important for firefighters to analyze and make decisions based on the audiovisual and possibly tactile information they collect. Such a scenario cannot leverage the current cloud computing paradigm, which has significant latency issues. That is why it is important to study and design edge computing paradigms to address the goals of these scenarios and use cases since they can be capable of fulfilling latency and reliability requirements. In this situation, this work looks into how computing at Edge Micro Data Centers (EMDC) can be used to improve how fires are predicted and managed. We propose a novel three-stage architecture. The initial stage focuses on prediction and classification of the fire occurrence based on available sensor data at the EMDC, whereas the second stage deals with the fire occurrence confirmation using a convolutional neural network (CNN) classification model. After the fire occurrence has been confirmed, the final stage notifies the tenants and streams 360-degree monitoring video to the nearby fire station after processing at EMDC. The results showed that the proposed architecture can realize firefighting services with low latency. to the best of the authors' knowledge, this is the first work studying and experimentally evaluating this communication scenario by also involving prediction via intelligence.