Transportation agencies rely on specialized inertial profilers to measure pavement roughness. However, the high cost and operational requirements of these systems limit the frequency and coverage of network-level condition assessments. This study presents a low-cost pavement monitoring framework that integrates smartphone sensing, physics-informed neural networks (PINNs), and computer vision for scalable roadway condition evaluation. Implemented as a React Native application, the system collects synchronized accelerometer, gyroscope, GPS, and roadway video data using a vehicle-mounted smartphone during routine travel. To estimate pavement roughness, the PINN predicts the International Roughness Index (IRI) directly from smartphone measurements while incorporating standardized quarter-car dynamics through a physics-based loss function. By embedding the industry-standard Golden Car model, the framework constrains predictions to remain consistent with the vehicle-pavement interactions underlying conventional IRI measurements. Model performance was evaluated using a leakage-free validation strategy in which entire roadway links were excluded from training, providing a rigorous assessment of generalization to previously unseen road segments. At the agency reporting scale, the proposed framework achieved an R² of 0.76 and a mean absolute error of 13.9 in./mile for IRI estimation. To complement roughness assessment, a YOLOv8-based distress detection model identified cracks, potholes, and surface deformations from roadway video with a mAP@0.5 of 97.1%. Roughness estimates and detected distresses were integrated within a GIS-based web platform, enabling synchronized geospatial visualization of pavement conditions across roadway networks. Results demonstrate the potential of low-cost mobile sensing and physics-guided AI to support scalable pavement monitoring and infrastructure management.