VerifyPK - AI Fact-Checking for Pakistan

VerifyPK uses RAG and LLMs to instantly fact-check viral news, WhatsApp claims, and headlines - returning a True/False/Unverified verdict with trusted Pakistani and global sources in seconds.

Comment

VerifyPK - AI Fact-Checking for Pakistan

Inspiration

VerifyPK was inspired by the growing spread of misinformation in Pakistan through social media, WhatsApp forwards, and online news platforms. During major events such as elections, health crises, and economic announcements, false information often spreads faster than verified facts. Many people lack access to quick and reliable fact-checking tools, making it difficult to distinguish truth from misinformation. We wanted to build an AI-powered solution that helps citizens verify claims instantly and make informed decisions based on credible evidence.

What it does

VerifyPK allows users to verify claims, headlines, social media posts, WhatsApp messages, and screenshots in seconds.

The platform:

  • Fact-checks user-submitted claims using real-time web search and fact-check databases.
  • Analyzes WhatsApp forwards for common misinformation patterns and red flags.
  • Extracts text from screenshots using OCR and verifies the extracted content.
  • Generates verdicts such as True, False, Misleading, or Unverified.
  • Provides Trust Scores, confidence levels, explanations, and supporting evidence.
  • Supports English, Roman Urdu, and Urdu script summaries.
  • Generates downloadable PDF fact-check reports for sharing and documentation.

How we built it

We built VerifyPK using a combination of AI, information retrieval, and fact-checking technologies.

Tech Stack

  • Python
  • Streamlit for the web interface
  • Groq (Llama 3.3 70B) for AI reasoning
  • Tavily Search API for real-time web search
  • Google Fact Check API for professional fact-check sources
  • Tesseract OCR for screenshot text extraction
  • FPDF for report generation

Workflow

  1. A user submits a claim, screenshot, or WhatsApp message.
  2. The system retrieves relevant information from trusted online sources.
  3. Existing fact-check databases are searched for prior investigations.
  4. OCR extracts text from uploaded screenshots when needed.
  5. The AI model analyzes only the retrieved evidence and generates a verdict.
  6. Results are presented with explanations, credibility scores, supporting sources, and Urdu summaries.
  7. Users can download a PDF report or share the results.

Challenges we ran into

One of the biggest challenges was ensuring that AI-generated responses remained factual and did not hallucinate information. We addressed this by grounding the model's reasoning in retrieved evidence from trusted sources.

Another challenge was handling multilingual content, particularly Urdu text commonly found in social media posts and WhatsApp forwards. Integrating OCR and multilingual AI support required additional testing and optimization.

We also faced challenges in balancing speed and accuracy. Users expect instant results, so we optimized API calls and retrieval workflows to keep response times low while maintaining quality.

Finally, determining source credibility was difficult because not all websites are equally trustworthy. To address this, we implemented a credibility scoring system that evaluates sources based on domain reputation.

Accomplishments that we're proud of

  • Built a complete AI-powered fact-checking platform tailored for Pakistan.
  • Successfully integrated multiple APIs into a single seamless workflow.
  • Developed a WhatsApp misinformation detector that identifies suspicious forwarding patterns.
  • Added screenshot fact-checking using OCR technology.
  • Implemented trust scores and source credibility ratings for greater transparency.
  • Created multilingual support with English, Roman Urdu, and Urdu summaries.
  • Generated professional PDF reports for fact-check results.
  • Designed a modern and user-friendly interface optimized for accessibility.

What we learned

Through this project, we gained practical experience in:

  • Retrieval-Augmented Generation (RAG) concepts
  • Information retrieval and evidence-based AI systems
  • Prompt engineering and structured AI outputs
  • API integration and workflow orchestration
  • OCR and document processing
  • Source credibility assessment
  • Building explainable and trustworthy AI applications
  • Developing production-ready Streamlit applications

Most importantly, we learned that transparency and explainability are essential when building AI systems that influence how people evaluate information.

What's next for VerifyPK - AI Fact-Checking for Pakistan

Our future roadmap includes:

  • Integrating VerifyPK directly with WhatsApp for one-click fact-checking.
  • Expanding support for Urdu voice input and voice-based fact verification.
  • Building a browser extension that verifies claims while users browse websites and social media.
  • Integrating additional Pakistani fact-checking organizations and government data sources.
  • Enhancing source credibility analysis using machine learning techniques.
  • Developing real-time misinformation trend monitoring across Pakistan.
  • Deploying the platform at a larger scale to make fact-checking accessible to millions of users.

Our long-term vision is to make VerifyPK the most trusted AI-powered fact-checking platform in Pakistan and help create a more informed digital society. 🇵🇰🔍

Built With

Share this project:

Updates