Author(s)
Prof. Sayali Gaikwad, Alfiya Shaikh, Aishwarya Jadhav, Parchi Khenget
- ISSN (P): 3139-8464
- Manuscript ID: 141068
- Volume: 2
- Issue: 10
- Pages: 288–293
Subject Area: Computer Science and IT
Abstract
Advances in generative adversarial networks and diffusion models have made synthetic face and voice media highly realistic, creating serious risks for misinformation, fraud, and identity abuse. This paper presents an AI-powered deepfake detection tool delivered as a web application. The system combines a React frontend, a FastAPI backend, a PostgreSQL database, an agent orchestrator that coordinates the analysis workflow, an isolated Docker sandbox for executing untrusted media through the analysis code, and a set of machine learning models that score media for manipulation. Users upload an image or video and receive an authenticity verdict with a confidence score. We describe the architecture, the detection pipeline, and the security rationale for sandboxed execution, and outline an evaluation protocol based on public benchmarks. The modular design allows detection models to be replaced or added without changing the user-facing application.