**VeriMedia: A Spatial-Temporal Deep Learning Platform for Video Deepfake Detection and Forensic Verification**

The rapid advancement of generative artificial intelligence has made it increasingly difficult to distinguish authentic digital videos from synthetically worked content, commonly known as deepfakes. These tweaks pose noticeable obstacles to digital media authenticity, cybersecurity, privacy and the reliability of visual proof, in real-world scenarios. In fact, standard deepfake detection approaches mostly depend on spatial attributes pulled from individual video frames, making them vulnerable to advanced generative procedures, video compression, and post-processing operations, in many respects. In digital forensic inquiries, and the lack of interpretability in existing deep learning models is limiting their applicability. To tackle these issues, what this paper suggests is VeriMedia, a spatial-temporal deep learning platform for video deepfake detection and forensic verification, in actual application. And thus, to sharpen model transparency and forensic interpretability, Gradient-weighted Class Activation Mapping (Grad-CAM) is worked in to make visual saliency heatmaps accentuating potentially controlled facial regions, to a significant degree. The put forward architecture is blending Multi-Task Cascaded Convolutional Networks (MTCNN) for facial detection, landmark alignment, and preprocessing, followed by a ResNeXt-50 backbone for pulling spatial feature representations from individual frames. These traits are later handled employing a Bidirectional Long Short-Term Memory (Bi-LSTM) network to encapsulate temporal inconsistencies, covering unnatural facial movements, inter-frame flickering, and irregular transitions that may show synthetic tweak. The suggested system is mainly intended to be checked employing benchmark datasets, covering FaceForensics++ and the Deepfake Detection Challenge (DFDC), under the distinct video compression conditions to assess detection accuracy, generalization and robustness, broadly speaking. The platform is currently designed with a modular architecture including a FastAPI-based asynchronous backend, Redis-powered background task processing, and an interactive React-based forensic dashboard, and is supporting video ingestion, frame-level examination, manipulation probability estimation, temporal anomaly visualization, and automated forensic report generation, on the whole. The platform seeks to support forensic analysts, journalists, and digital investigators in assessing media authenticity and strengthening trust in digital visual content, to a significant degree. Indeed, temporal sequence modelling by combining spatial feature extraction, and explainable artificial intelligence, VeriMedia aims to offer a broad and interpretable method to video deepfake detection, broadly speaking.