🛡️ KAVACH: Privacy-First AI Interaction

🔐 "Your Personal Redaction Shield Before Talking to AI"

---

📖 Project Overview

KAVACH is a powerful, privacy-centric text redaction and sanitation system built to shield sensitive user data from exposure to Large Language Models (LLMs) 🤖.

It acts as a smart intermediary — identifying and redacting Personally Identifiable Information (PII) from your text so you can interact with AI without compromising your privacy 🔏.

🎯 Engineered using a multi-layered AI stack, KAVACH achieves over 89% accuracy in PII detection, making it a reliable privacy protector in digital communication.

---

✨ Features

• 🔍 Intelligent PII Redaction
  Detects and replaces sensitive entities like names, phone numbers, addresses, emails, and more.

• 🇮🇳 Support for Indian PII
  Recognizes Aadhaar, PAN, and other India-specific identifiers.

• 🧠 Contextual Data Protection
  Uses NER models to identify sensitive data even without clear patterns (like uncommon names or cities).

• 🏷️ Text Sensitivity Classification
  Labels input as confidential, personal, or public using LLM-based text classification.

• 🔗 Seamless LLM Integration
  Ensures only redacted & safe text is passed to AI models.

---

🧬 How It Works: Layered Protection System

KAVACH uses a 3-layer defense strategy 🧱 for robust privacy:

1️⃣ Pattern-Based Scanning

🛠️ Utilizes Presidio to scan and tag PII with recognizable formats:

• Phone → [PHONE_NUMBER]
• Email → [EMAIL_ADDRESS]
• Aadhaar, PAN → [ID_NUMBER]

2️⃣ Contextual Detection (NER)

🤖 Employs ai4bharat/IndicNER from HuggingFace for Named Entity Recognition:

• Detects names, locations, organizations, etc. without pattern reliance.

3️⃣ Sensitivity Classification

🧠 Leverages facebook/bart-large-mnli to assign sensitivity tags:

• Confidential 🛡️
• Personal 🙋‍♂️
• Public 🌍

Only after these steps is the sanitized text released for AI use.

---

💻 Local Setup & Run

Want to try it out? Here’s how you can run it locally 🧪:

🔧 Prerequisites

• Python 3.8+
• pip (Python package installer)
• Git

🛠️ Setup Steps

📅 Step 1: Clone the Repository

git clone https://github.com/thechiranjeevvyas/Kavach.git
cd RAKSHak

🧱 Step 2: Create and Activate a Virtual Environment

Create a virtual environment:

python -m venv venv

Activate it:

• macOS / Linux:

source venv/bin/activate

• Windows (Command Prompt):

.\venv\Scripts\activate

• Windows (PowerShell):

.\venv\Scripts\Activate.ps1

✅ You’ll see (venv) in your terminal prompt when activated.

📦 Step 3: Install Dependencies

Install required Python packages:

pip install -r requirements.txt

Sample requirements.txt content:

streamlit
presidio-analyzer
python-dotenv
transformers
torch
groq
sentencepiece
accelerate

🔑 Step 4: Set Up Groq API Key

To connect with the LLM securely:

Create a .streamlit folder:

mkdir .streamlit

Inside it, create secrets.toml:

GROQ_API_KEY = "your_groq_api_key_here"

🔐 Replace with your real API key from Groq Console.

▶️ Step 5: Run the Streamlit App

streamlit run main2.py

🌐 Opens in your browser at: http://localhost:8501

---

🧪 Testing Your Application

🔹 Comprehensive Input

Urgent internal memo: This document contains highly confidential information. Patient Anjali Sharma (DOB: 15/03/1988) visited Apollo Hospital on 2024-06-20 for follow-up. Her unique patient ID is PX7890123. The physician Dr. Rajesh Kumar (Mobile: +919876543210) noted her AADHAAR number 9876 5432 1098. She works for TechSolutions India. Employee ID E12ABU5678 is assigned to Mr. Vikram Singh, a senior analyst at State Bank of India. His PAN is ABCDE1234F and email is vikram.singh@examplebank.com. We also received a query from a Ministry of Defense official regarding vehicle registration DL01CD1234 for a new project located near the Air Force Station, Hindon. Please ensure all PII and sensitive project details are redacted before sharing any summaries. Contact our legal department at legal@techsolutions.com for further clarification. Voting ID of Ms. Priya Patel: ABC1234567. Passport No. K1234567.

🔹 Short Input: Personal Info

Could you please help me confirm my identity? My full name is Sarah Miller, and my private phone number is +1-202-555-0100. Thanks.

🔹 Short Input: Government Sensitive Info

Please redact details from this secure document. The official government ID for the operation is GVT-SEC-98765, linked to agent E78XYZ4321.

---

🤝 Contributing

• Fork the repository
• Create a new branch
• Commit changes
• Submit a pull request

For significant changes, open an issue to start a discussion.

---