All projects
Project 04 · AI · Web App
Deepfake Detection
A full-stack AI-powered web application that detects deepfakes and classifies sensitive content in uploaded images. It integrates Google Cloud Vision API via Supabase Edge Functions for server-side inference, keeping credentials secure while returning per-category confidence scores to a React frontend with drag-and-drop upload and a visual results dashboard.
ReactTypeScriptVite Google Cloud Vision API Supabase Edge FunctionsNode.js
Real-time AI classification
Google Cloud Vision API runs inference on uploaded images and returns confidence scores for multiple harm categories including deepfakes, adult content, violent material, and harassment. Results render instantly on the dashboard alongside visual confidence bars.
Drag-and-drop interface
A modular React + TypeScript frontend supports drag-and-drop or click-to-upload image input. The results dashboard displays per-category scores, a final verdict (safe / flagged), and a confidence breakdown — all rendered without page reload.
Smart confidence threshold
A 0.6 confidence threshold filters out low-signal detections before flagging content as harmful, significantly reducing false positives. The threshold is configurable in backend logic without frontend changes, making it easy to tune for different use cases.
Secure server-side inference
All API calls to Google Cloud Vision are routed through Supabase Edge Functions, ensuring credentials are never exposed to the browser. The frontend only receives processed classification results — the raw Vision API response stays entirely server-side.