Pujan-Dev/test

AI Text Detector — FastAPI App Documentation

About

This FastAPI application predicts whether a given text is AI-generated or human-written based on a GPT-2 model.
It measures the perplexity of the input text to make the decision.

You can access the live version here:
Public App Link

---

API Endpoints

GET /

• Description: Welcome route showing basic API information.
• Response:

  {
    "message": "FastAPI API is up.",
    "try": "/docs to test the API.",
    "status": "OK"
  }
  

GET /health

• Description: Health check to verify if the server is running.
• Response:

  {
    "status": "ok"
  }
  

POST /analyze

• Description: Analyze input text to determine if it is AI-generated or human-written.
• Authorization: Requires Bearer Token.
• Request Body:

  {
    "text": "Enter your text here"
  }
  

• Response:

  {
    "result": "AI-generated" | "Probably AI-generated" | "Human-written",
    "perplexity": 47.35
  }
  

• Important:
  • Input text must contain at least two words.
  • Without a valid token, this endpoint will not work.

---

Authorization Guide

• Open /docs (Swagger UI).
• Click the "Authorize" button at the top right.
• Paste your Bearer Token into the Authorization field.
• After authorization, you can use the /analyze endpoint.

Example Authorization header:

Authorization: Bearer YOUR_SECRET_TOKEN

---

How the AI Detection Works

• The app loads a fine-tuned GPT-2 model at startup.
• It calculates the perplexity of the input text:
  • Perplexity less than 60 → AI-generated
  • Perplexity between 60 and 80 → Probably AI-generated
  • Perplexity greater than 80 → Human-written

Higher perplexity indicates the text is harder for the model to predict, usually meaning it is human-written.

---

How to Run Locally

1. Clone the repository:

   git clone https://huggingface.co/spaces/can-org/canspace
   cd canspace
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Start the server:

   uvicorn app:app --reload
   

4. Open the API docs at:
   http://localhost:8000/docs

Quick Reminders

• Always authorize first in /docs before using /analyze.
• Text input must contain more than two words.
• Public live app is available at: https://can-org-canspace.hf.space