VerifiAI

VerifiAI explores how people can assess AI-generated content more responsibly by combining source review, provenance context, and careful product design.

Santosh Poudel, founder of VerifiAI, in a portrait for the project

About the project

Review content with context

VerifiAI explores ways to assess AI-generated and manipulated digital content using source review and provenance context. The project is designed around the idea that a detector result is one signal to consider, not a substitute for checking where content came from.

A missing provenance record does not prove that content is false or AI-generated. Platform uploads can remove metadata, and many creators do not attach credentials.

Be clear about what detection can establish

AI detection can be uncertain. A tool's result should not be treated as conclusive proof of authorship, authenticity, or truth. Source context and other available evidence matter when evaluating digital content.

VerifiAI is an ongoing project founded by Santosh Poudel. Its project page documents the goals and limitations, while the separate product site describes the current product.

Highlights

  • Practical AI content review workflow
  • Focus on transparency and limitations
  • Built for creators, educators, and teams evaluating digital trust

Tech stack and project areas

  • Next.js
  • AI workflows
  • Content verification
  • Web tooling

Read why I built VerifiAI and how provenance relates to AI content.

Frequently asked questions

What is VerifiAI?

VerifiAI is a project founded by Santosh Poudel that explores assessing AI-generated and manipulated digital content using source review and provenance context.

Can a VerifiAI result prove who created content?

No. AI detection can be uncertain, and a detector result is one signal rather than conclusive proof of authorship, authenticity, or truth.

Does this page explain VerifiAI's detection implementation?

No. This portfolio page does not document the detector's model, training data, thresholds, or evaluation results, so those details are not inferred here.