Deepfake Detection System

Full-stack AI platform with a 4-agent orchestration system for deepfake media analysis, powered by ResNet50/EfficientNet trained on FaceForensics++.

A full-stack platform that detects AI-generated and manipulated media, built around a multi-agent pipeline.

Architecture

The system orchestrates four agents, each with a focused job:

Perception — ingests and normalizes media (frames, audio) via FFmpeg.
Detection — runs the ML classifiers over extracted frames.
Compression — handles efficient storage and streaming of analyzed assets.
Cognitive — aggregates signals into a final, explainable verdict.

The models

Trained ResNet50 and EfficientNet classifiers on the FaceForensics++ dataset — 16GB and 7000+ samples — achieving high accuracy in distinguishing genuine footage from AI-generated content.

Stack

Next.js · React · TypeScript · Node.js · Express · MongoDB · FFmpeg · Tailwind CSS