Unmask AI

Unmask AI is an AI-powered system that exposes fake news, bias, and manipulation

Comment

Inspiration

Misinformation spreads faster than ever, especially through images and short-form content where context is lost. News screenshots, manipulated headlines, and biased narratives are shared widely without verification. Unmask AI was inspired by the need for a fast, accessible tool that helps people understand whether a piece of information is likely fake and what ideological bias it may carry, instead of blindly trusting or rejecting it.

What it does

Unmask AI analyzes images or text containing news-related content and:

  • Extracts text using Optical Character Recognition (OCR)

  • Analyzes the content with AI to estimate:

    • The probability of being fake or misleading
    • The presence and direction of ideological bias (e.g. conservative, liberal, neutral)

  • Provides a structured explanation to help users understand why the content may be unreliable

The goal is not censorship, but critical thinking and transparency.

How we built it

Unmask AI was built as a modular AI pipeline:

  • Image text extraction using OCR (PaddleOCR)

  • Text cleaning and normalization

  • AI-based content analysis using a language model fine-tuned or prompted for misinformation detection

  • Backend logic developed in Node.js

  • Clear JSON-based outputs to make results explainable and easy to integrate into future platforms

The system was designed to be scalable and adaptable to different datasets and languages.

Challenges we ran into

  • Handling noisy or low-quality text extracted from images

  • Differentiating fake news vs. biased but factual content

  • Designing prompts and outputs that are explainable, not just binary decisions

  • Limited access to high-quality labeled datasets for ideological bias classification

Accomplishments that we're proud of

  • Building a complete end-to-end system in a limited time

  • Successfully integrating OCR with AI-based misinformation analysis

  • Designing outputs that prioritize transparency and reasoning

  • Creating a project with real-world relevance and social impact

What we learned

  • Fake news detection is more about context and framing than simple keywords

  • Bias exists on a spectrum and must be handled carefully

  • Explainability is essential for trust in AI systems

  • Clean data and preprocessing are as important as the model itself

What's next for Unmask AI

  • Expanding multilingual support

  • Improving fine-tuning with larger, more diverse datasets

  • Adding source verification and cross-referencing with trusted media

  • Deploying Unmask AI as a web and mobile application

  • Exploring partnerships with educators, journalists, and fact-checking organizations

Built With

Share this project:

Updates