1 of 8

"We are building a tool to combat AI misinformation. It’s a mobile app or browser extension that analyzes media clips and provides a 'manipulation score' using deep learning models to identify synthetic speech or video."

2 of 8

Project goal

  • We want to build web extension/mobile app that allow user to upload a video and return feedback based on how real the video is, specifically telling user if this video has been altered by AI or not.

3 of 8

Progressive goals�

Week 1: Setup

Goal: Everyone hasdevelopment env ready.

Tasks:

Download a standard dataset (e.g., FaceForensics++ subset).

Decides on a pre-trained deepfake detection model from GitHub, setup git.

Week 1

Week 2

Week 3

Week 4

Week 5

Week 6

Week 7

Week 8

Week 9

Week 10

Week 11

Week 12

Week 2: The Pipeline Script

Goal: A script can process a video file.

Tasks:

Write code to extract frames from a video.

Write code to crop faces from those frames.

Week 3: Output prediction

Goal: The script outputs a prediction.

Tasks:

Feed cropped faces into the pre-trained model.

Aggregate frame scores into a single video score.

Deliverable: A script you run in the terminal: python detect.py video.mp4 -> Result: FAKE, score:12

-> Result: Real, score:87

Week 4: Create server

Goal: API is running locally.

Tasks:

Create a simple server.

Create an endpoint that accepts a file upload and saves it to a temp folder.

Week 5: Integration

Goal: The API runs the AI.

Tasks:

Connect the endpoint to the Python script from Week 3.

Return the script's output as JSON: {"prediction": "Fake", "confidence": 0.98}.

Week 6: Rough Interface

Goal: A basic frontend talks to the backend.

Tasks:

Build a minimal frontend with one button: "Upload Video".

Display the raw JSON response on the screen.

Week 7: Speed Optimization

Goal: Reduce waiting time.

Tasks:

Optimize the pipeline (e.g., check only 1 frame every second instead of every frame).

Implement a "Loading..." spinner in the UI so the user knows it's working.

Week 8: UI Polish

Goal: Clear visual feedback.

Tasks:

Replace raw JSON text with clear UI elements.

Add a progress bar for the upload.

Week 9: Full System Test

Goal: Verify accuracy andstability.

Tasks:

Test with new videosthe AI hasn't seenbefore.

Fix any critical bugsfound during testing.

Week 10: Documentation

Goal: Project is hand-offready.

Tasks:

Write the FinalReport.

Clean up codecomments.

Create a READMEwith installationinstructions.

Week 11:  Buffer week

Goal: finishes any unfinished task from previous week

Tasks: Decide as time approach

Week 12:  Submission &Demo

Goal: Final deliverables.

4 of 8

Roles

  1. ML model researcher: Focuses on the Python ML model. Finds pre-trained models, writes the inference script, and improves accuracy.
  2. Backend developer: Setup the AI model in a web server. Handles video processing and sends results back to the frontend app.
  3. Frontend developer: Builds the mobile app (React?) or Web Interface. Focuses on UI/UX design for uploading videos and viewing results.
  4. QA: Finds datasets (Real vs. Fake videos). Later, they act as the "attacker," generating new deepfakes to try and fool Student A's model. Write unit test and integration test

5 of 8

Tech stack

  • Pytorch
  • OpenCV
  • FastAPI/Node.js/Express.js
  • React/Streamlit
  • Sqlite

6 of 8

Github links:

7 of 8

related papers

8 of 8

Architecture diagram