Trinetra is a hybrid deepfake forensic analyzer designed to detect media manipulation through a multi-layered technical approach. To overcome the limitations of standard CNNs, the system implements a local hybrid pipeline combining EfficientNet-B4 for frame-by-frame spatial feature extraction with a Long Short-Term Memory (LSTM) layer to track temporal inconsistencies and glitches across video sequences. To move beyond simple binary classification, the architecture integrates Explainable AI (XAI) techniques, including Grad-CAM heatmaps for spatial localization of manipulations, Error Level Analysis (ELA) to identify compression noise residuals, and landmark jitter metrics to detect unnatural geometric facial movements. The ecosystem features a Gradio-based web application, a Chrome extension powered by a FastAPI backend, and a WhatsApp bot utilizing the Meta Cloud API. For media that exceeds local hardware capabilities, the system includes a cloud fallback mechanism via the Reality Defender API. This comprehensive framework aims to provide both a confidence score and visual forensic evidence to justify its classification results.