Privasee

🧠 About the Project

🌟 Inspiration

Privacy policies are long, confusing, and written in legal jargon that most users skip over.
We created PrivaSee to make digital privacy transparent and accessible — helping people instantly understand what they’re agreeing to before clicking “Accept.”

🧩 What We Built

PrivaSee is an AI-powered Chrome extension that analyzes any privacy policy page in seconds.

It uses:

Natural Language Processing (NLP) with spaCy and custom heuristics to identify key clauses.
A Flask backend to process and classify text into categories like Data Collection, Sharing, and User Rights.
A React frontend that displays a personalized trust score, an overview, and conflict highlights based on user privacy preferences.
Ngrok tunneling to securely expose the local backend for testing and cross-device access.

Mathematically, each policy section’s trust score is computed as:

$$ \text{score}_{category} = \frac{\text{matched_safe_criteria}}{\text{total_criteria}} \times 100 $$

🧰 How We Built It

Frontend: React + Vite + Chrome APIs for extension features.
Backend: Python (Flask) with spaCy, heuristic scoring, and CORS setup for local testing.
Integration: Service worker fetches user preferences from chrome.storage.local and sends them to the backend via fetch() calls to the ngrok HTTPS endpoint.
AI Layer: NLP pipeline extracts entities, key phrases, and risky terms, generating both a structured JSON response and a natural-language summary.

💡 What We Learned

How to build a full-stack Chrome extension that communicates securely with an AI backend.
Techniques for scoring and classifying unstructured legal text using NLP.
How to handle CORS, mixed content, and HTTPS tunneling through ngrok.
The importance of aligning UI design with trust and readability principles.

⚙️ Challenges We Faced

Managing CORS issues between the extension and the backend API.
Debugging service worker fetch calls while testing locally and via ngrok.
Ensuring consistent LLM responses and refining heuristic scoring for accuracy.
Balancing real-time responsiveness with API latency constraints.