Skip to main content

​
The Mathematics of Misinformation

Every minute, 500 hours of video are uploaded to YouTube. That’s 720,000 hours per day. Meanwhile, professional fact-checkers can review maybe 10-20 claims per day. The math simply doesn’t work. Traditional automated tools only analyze text, missing on-screen graphics, visual demonstrations, and “white coat” authority signals. VerityNgn (Verity Engine) is an open-source system that doesn’t just read transcripts—it watches, listens, and verifies the truth.

​
What is VerityNgn?

VerityNgn is a multimodal analysis engine designed to fact-check video content at scale. It moves beyond simple RAG pipelines to provide a comprehensive verification suite:
  • 🎥 Multimodal Analysis: Sampling at 1 FPS to capture OCR, visual context, and audio nuances.
  • 🔍 Evidence Verification: Cross-referencing claims with scientific databases and credible web sources.
  • 🕵️ Counter-Intelligence: Actively hunting for contradictory reviews and promotional press releases.
  • 📊 Probabilistic Assessment: Delivering nuanced verdicts based on Bayesian aggregation of evidence.

​
Why Open Source?

Misinformation is too big a problem to solve behind closed doors. We’ve released VerityNgn under the Apache 2.0 license to ensure the methodology is transparent, the code is auditable, and the community can contribute to a more truthful internet. Star us on GitHub | Read the release announcement on Substack

Public Demo

Try VerityNgn right now in your browser.

Quickstart

Get VerityNgn running locally in under 5 minutes.

Architecture

Deep dive into the 1M token context window and intelligent segmentation.