👋 Welcome to My GitHub

I focus on building secure, privacy-aware tools powered by AI.
Here, code does the talking — no overshare, just solutions.

---

🚫 NOovershare

NOovershare is a privacy-first platform designed to analyze and sanitize images before they’re shared online. It detects risks like personal data, sensitive content, or inappropriate visuals — and helps users prevent unintentional oversharing.

🔐 Key Features

• 🎯 Object Detection — Identifies faces, documents, credit cards using DETR
• 🧾 Text Extraction — Pulls text (like addresses or phone numbers) using EasyOCR
• 🚫 NSFW Detection — Flags explicit content via AdamCodd
• 🧠 Risk Summarization — Uses LLaMA to provide natural-language risk reports
• 🖼️ Editing Tools — Crop or blur selected parts of the image before sharing
• 🔒 End-to-End Encryption — AES-128 (Fernet) with PBKDF2-HMAC + SHA256
• ⚡ Groq Integration — Hardware acceleration for model inference

---

🛠️ How to Run Locally

1. Clone the Repository

git clone https://github.com/yourusername/noovershare.git
cd noovershare

2. Set Up a Virtual Environment

python -m venv venv
source venv/bin/activate  # or venv\Scripts\activate on Windows

3. Install Dependencies

pip install -r requirements.txt

You can adjust versions if you know your project uses different ones. If you want to generate this file from your current environment, use:

pip freeze > requirements.txt

4. Enable Groq Acceleration

   Create a .env file in the root directory:

    USE_GROQ=True
    GROQ_API_KEY=your_actual_key_here

6. Start the FastAPI Backend

uvicorn app.main:app --reload

6. Launch the Frontend

Open index.html in your browser to access the image upload, crop, and risk analysis UI. 📎 Tech Stack

Frontend: HTML, CSS, JavaScript, Cropper.js

Backend: Python, FastAPI, Uvicorn

Models: DETR, EasyOCR, AdamCodd, LLaMA

Crypto: Fernet (AES-128), PBKDF2HMAC, SHA256

Processing: OpenCV, PIL

Acceleration: Groq hardware (optional)

🤖 Project Status

Prototype live with all major components integrated. Supports:

Drag & drop image uploads

Region-specific blurring

Risk summaries

Encrypted downloads

Further work planned on:

False-positive tuning in NSFW detection

Improved OCR on low-res images

Dockerized deployment