DEV Community

thetruecopy
thetruecopy

Posted on

Building an AI Text Detector with RoBERTa & Burstiness Analysis

Hey everyone! I just deployed Veridraft AI Detector Pro, a Streamlit app designed to catch AI-generated text without relying solely on basic word probabilities.

Key Features

  • Global Context Scan: Evaluates overall document context alongside individual sentence windowing.
  • Burstiness Scoring: Measures sentence length variance (CV = std_dev / mean) to catch uniform, robotic structures.
  • Normalized Parsing: Cleans hidden characters in PDF/DOCX uploads so outputs match raw text.
  • Tech Stack: Python, Streamlit Cloud, Hugging Face (Hello-SimpleAI/chatgpt-detector-roberta), PyTorch, PyPDF.

🔗 Live Demo: https://thetruecopy-veridraft-detector-app-thguvb.streamlit.app

Would love for you to test it out and drop any feedback or edge cases!

Top comments (0)