Tiktok TechJam 25 - Thunderbolt

Privacy Meets AI: Building a Safer Digital Future

AI-powered visual privacy protection — detect, tag, and mask sensitive information from your images before they leak.

---

Problem Statement

As AI integrates deeper into our daily lives, privacy risks grow more pressing.
From hidden location cues in photos to unauthorized data leaks, users face challenges in controlling how their data is exposed.

Thunderbolt addresses this problem by combining Generative AI with privacy-first design:

1. Making AI Systems Resilient Against Attacks & Misuse

• Local On-Device Processing — processes images locally whenever possible, reducing reliance on cloud inference and minimizing security risks.
• No Data Retention — images are not stored after detection, preventing re-identification or unauthorized reuse.
• Robust Detection Pipeline — multi-layered detection flags potentially malicious or privacy-invasive inputs early.
• Privacy-Preserving Defaults — sensitive feature detection and masking are always enabled by default.

2. Using AI to Protect Users' Privacy

• Sensitive Feature Detection — identifies faces, license plates, signs, and other geo-inferable elements automatically.
• Danger-Level Scoring — quantifies image sensitivity with a percentage-based privacy risk score.
• Automatic Privacy Protections — blurs or masks sensitive areas to prevent location leakage and identity exposure.
• Explainable Insights — shows users which features were flagged, why, and how they were masked.

---

Features

Risk Detection Dashboard

• Upload an image and instantly analyze potential privacy risks.
• Detect identifying cues like:
  • Faces
  • Road signs
  • Building structures
  • License plates
  • Location-sensitive elements

Dynamic Danger Level Gauge

• AI-powered scoring system highlights the risk level of each uploaded image.

• Color-coded gauge:

  • 🟢 Low Risk
  • 🟡 Medium Risk
  • 🔴 High Risk

  

Automated Privacy Protection

• Blur or mask sensitive regions in images (e.g., plates, street signs).
• Designed to prevent geo-location inference and identity exposure.

Interactive Visualizations

• Circular gauge: Displays the danger percentage intuitively.
• Bar chart: Highlights the count of detected sensitive features.

---

System Architecture

graph TD
  A[User Uploads Image] --> B[On-device Processing]
  B --> C[AI-powered Detection]
  C --> D[Privacy Risk Scoring]
  C --> E[Feature Tagging]
  D --> F[Danger Level Gauge]
  E --> G[Bar Charts + Insights]
  F --> H[Masked Image Output]

Loading

• Frontend: React (expo mini-app framework)
• AI Engine: Integrates generative AI models for detection and inference
• Privacy Layer: Image processing techniques to blur/mask sensitive regions

---

Tech Stack

Frameworks & Libraries

• React-native (Expo) — UI/UX development
• TypeScript — Strong typing for safer, cleaner code
• CSS Modules — Encapsulated styling for components
• Generative AI APIs — Model inference and location-based sensitivity scoring

Development Tools

• Visual Studio Code (VSCode) — for project development, debugging, and code management.
• Expo Mini-App Framework — frontend framework for seamless mobile-native UI development.
• Node.js & npm — used for dependency management and running the development server.
• Git & GitHub — version control and collaborative development.
• Jupyter / Colab — quick experimentation for AI detection models.

Key Components

File	Description
Thunderbolt.tsx	Main dashboard & detection logic
HomeGpt.tsx	Welcome page with result visualization
Query.tsx	Handles uploads, previews, circular gauges, bar charts
App.tsx	Entry point for Expo mini-app
Router.tsx	Manages page routing within the app

---

🔗 APIs & Assets Used

AI Models

• Open-source detection models: YOLOv8 / CLIP-based analysis
• Generative AI: OpenAI for location-risk predictions
• Roboflow OCR API: for text extraction from images to detect location-revealing information

Assets

• Custom icons:
  • homeIcon.png
  • boltIcon.png
• Brand-specific UI elements

---

How It Works

Step 1 — Upload & Local Preprocessing

• Images are uploaded from the user’s device.
• Metadata such as EXIF location tags is automatically stripped to prevent accidental leaks.

Step 2 — Analyze with AI-Powered Privacy Risk Detection

• Thunderbolt uses Roboflow OCR API to extract visible text and other sensitive from uploaded images. This helps flag sensitive location-revealing information such as:
  • Faces
  • Vehicle plates
  • Landmarks & building patterns
  • Road signs or street names

Step 3 — Privacy Risk Scoring

• The system quantifies privacy exposure using:
  • Danger-level scoring (0–100%)
  • Context-aware risk thresholds based on detected features.

Step 4 — Automated Privacy Protection

• High-risk areas are blurred, masked, or noise-injected dynamically.
• Output images are optimized for safe sharing online.

Step 5 — Explainable Insights

• Thunderbolt provides visual analytics:
  • A circular danger gauge for risk at a glance.
  • A bar chart showing how many sensitive features were detected.

---

Future Enhancements

1. On-device AI

• Process images entirely on-device for maximum privacy.

2. Generative Adversarial Masking

• Use AI to create context-preserving blurs for masked areas.

3. Social Media Integration

• Pre-scan images before sharing to ensure no sensitive data leaks.

4. Real-Time Video Privacy

• Extend detection and masking to live video streams.

---

Installation

Clone the repository

git clone https://github.com/<your-org>/thunderbolt

Navigate into the project

cd thunderbolt

Install dependencies

npm install

Start development server

npm start

---

📜 License

This project is released under the MIT License.
See the full license text here.

---