TruthLens: A Serverless Deepfake Detection Platform Leveraging MesoNet and AWS Cloud Services

Advancements in deepfake technology have escalated challenges in verifying digital media authenticity, posing risks to information integrity and public trust. This paper presents TruthLens, a scalable and efficient deepfake detection platform designed to identify manipulated visual content in real-time. The system leverages a lightweight CNN architecture, optimized for detecting subtle manipulation artifacts, and integrates it with a user-friendly web interface and serverless cloud infrastructure. The methodology involves curating a balanced dataset of real and fake images, applying data augmentation to intensify model robustness, and training the CNN to distinguish authentic from manipulated content. The platform employs a modular design, with a React-based frontend for seamless user interaction, a Flask API for real-time inference, and AWS services Lambda for computation, S3 for hosting, and DynamoDB for analytics to ensure scalability and cost-efficiency. Evaluation on diverse test set demonstrates the system's effectiveness in accurately classifying media, with low false positives to maintain user trust. Qualitative analysis confirms its ability to detect common deepfake artifacts, such as unnatural facial features. TruthLens advances digital forensics by offering a practical, accessible solution for media verification, with future enhancements planned for video support, explainable AI visualizations, and resilience against adversarial attacks. This work contributes to combating digital misinformation, supporting applications in journalism, content moderation, and digital security.