All projects
AI / MLResearch2026
Deepfake Webcam Detection
Real-time facial manipulation detection
Role / Solo — model and interface

- face classes
- 3
- webcam inference
- Realtime
- no cloud upload
- Local
The problem
Deepfake detectors are usually benchmark scripts. Demonstrating one convincingly needs a live, local interface that never uploads the user's camera feed.
Approach
- Trained a classifier across real faces, beauty/AR-filtered faces and AI-generated faces.
- Wrapped inference in a Python service and streamed frames from a local Next.js GUI.
- Kept all processing on-device so the demo works offline and handles sensitive video safely.
Result
- Working local demo with real-time webcam classification.
- Public repository with the training pipeline and the GUI.
Stack
- Python
- PyTorch
- OpenCV
- Next.js
- TypeScript