Deep Guard AI

DeepGuard AI — AI-Powered Deepfake Detection

Inspiration

With the rapid growth of generative AI, manipulated images and synthetic media are becoming increasingly realistic. Deepfakes can be misused to spread misinformation, impersonate individuals, and undermine trust in digital content. We wanted to build a practical tool that helps people analyze digital media using AI and better understand the risks associated with manipulated content.

What it does

DeepGuard AI is a web-based AI image detection application that analyzes submitted images and predicts whether they are AI-generated. It uses an AI detection model to produce a prediction and presents the result through a user-friendly web interface. The goal is to make AI-assisted media verification more accessible while encouraging users to interpret detection results responsibly.

How we built it

We built DeepGuard AI using a modern full-stack architecture:

Frontend: React, Vite, and TypeScript for the web interface.
Backend: Python and FastAPI for API endpoints and image-processing workflows.
AI/ML: ONNX Runtime with an ONNX-based AI image detection model.
Authentication and data integration: Supabase.
Deployment: Render for cloud hosting.

The image analysis pipeline preprocesses the input image, runs model inference, converts the output into an AI probability, and applies a decision threshold to produce the displayed classification.

Challenges we ran into

One of our biggest challenges was integrating the AI model with the web application. We also worked through API routing, frontend-backend communication, CORS configuration, authentication token handling, and cloud deployment issues. Ensuring that the model's output was interpreted correctly was another important challenge, because an incorrect label mapping could cause the application to display misleading results.

Accomplishments that we're proud of

Built and deployed a full-stack AI detection web application.
Integrated an ONNX-based AI image detector into a backend API.
Connected the frontend and backend into a unified analysis workflow.
Integrated Supabase authentication.
Made the application accessible through a live cloud deployment.
Created a foundation for expanding the system into a more comprehensive media verification platform.

What we learned

We gained practical experience in full-stack development, AI model integration, image preprocessing, API design, authentication, and cloud deployment. We also learned that AI detection requires careful testing and evaluation. A model's confidence score is not absolute proof, and responsible communication of limitations is essential when building tools for digital media verification.

What's next for DeepGuard AI

Our next steps are to improve detection reliability through broader model evaluation and benchmarking, add video frame and temporal analysis, explore audio deepfake detection, and introduce explainable AI visualizations. We also plan to investigate advanced forensic reports, background processing for larger files, and GPU-based inference to support more demanding workloads.

Live Project: https://deepguard-ai-isaac.onrender.com/