Showcase: Veridraft AI Detector Pro (Context-Aware Ensemble + PDF Parsing)

Show the Community!
1

Hey Streamlit Community! :waving_hand:

I’m sharing a project I’ve been building and tuning: Veridraft AI Detector Pro, a multi-layered AI text detection app built entirely with Streamlit and HuggingFace Transformers.

Traditional single-model classifiers often struggle with score dilution or file-parsing bugs. To fix this, I built a hybrid detection approach directly into the Streamlit workflow.

Key Features & Fixes:

  • :robot: Context-Aware Sliding Window: Evaluates target sentences alongside surrounding context to preserve semantic flow and eliminate probability dilution.

  • :bar_chart: Sentence Variation (Burstiness): Tracks structural sentence length variation to catch uniform AI writing patterns.

  • :page_facing_up: Normalized Document Parsing: Solved PDF formatting artifacts (hidden line breaks/tabs) so uploaded .pdf and .docx files return 1:1 identical scores to pasted text.

  • :magnifying_glass_tilted_left: Interactive Forensics: Color-coded sentence-by-sentence risk highlighting, live sidebar session stats, and downloadable forensic CSV reports.

Tech Stack: Streamlit, HuggingFace transformers (chatgpt-detector-roberta), pandas, numpy, pypdf, python-docx.

:link: Live App: https://thetruecopy-veridraft-detector-app-thguvb.streamlit.app/

:octopus: GitHub Repo: GitHub - thetruecopy/veridraft-detector · GitHub

Would love for the community to test it out, try to break it with edge cases, or share feedback on the UI and scoring logic!