Potholes are becoming a big problem for our road safety. This problem is leading to accidents and unfortunate incidents when drivers don’t steer past them quickly. The technology available today in the automobiles is not that effective and is not sufficient. Hence making it crucial to develop an efficient method to detect potholes to alert drivers in real time. An advanced pothole detection system can make a difference by improving road safety and lowering the accidents caused by these bumps, also seamlessly integrating with existing systems like ADAS. The solution proposes a pothole detection system that uses AI camera vision to detect potholes in real time. The AI vision cameras use YOLO algorithm to scan the road ahead to detect the potholes in real time based on visual cues so that the driver gets enough time to react to it. You Only Look Once (YOLO) is an AI algorithm which uses deep learning and is generally used for real-time object detection. It is designed for making quick detections, making it ideal for applications which require quick response from the user. It uses a bounding box, a box around the detected object to specify it. Also, YOLO can be fine-tuned with a custom dataset, enabling precise detection tailored to the problem. The YOLO algorithm, along with detecting the potholes, also estimates the size of the pothole and classifies it as big or small through visual cues. The bigger potholes are detected with a red bounding box and the smaller ones with a green bounding box. This acts as a form of visual real-time alert without distracting the driver from the road. The total number of potholes detected in the video is counted, and the road is classified as pothole-prone or safe based on predefined thresholds. The integration of the Mapbox API allows users to map already analysed road segments, offering a visual representation of pothole-prone areas. However, the current road classification is not performed in real time and can be manipulated by feeding altered video input. This limitation can be addressed by incorporating GPS functionality, enabling the system to operate in real time and accurately classify the road segments the user wants to analyse. The system was evaluated using three YOLO model variants: YOLO11-n, YOLO11-s, and YOLO11-m. The results show that YOLO11-M achieved the highest mean Average Precision (mAP) score of 0.804 but required the longest computational time of 6.100 hours. YOLO11-S had a mAP of 0.796 with a computational time of 1.005 hours, while YOLO11-N recorded a mAP of 0.788 and the shortest computational time of 0.710 hours. These results highlight a trade-off between detection accuracy and processing time, allowing for model selection based on application needs.