VideoFactChecker

Inspiration

In today's digital world, videos spread faster than facts. Platforms like YouTube, X (Twitter), and Instagram have become primary sources of news for millions of people. While this has made information more accessible, it has also created an environment where misinformation can spread rapidly and influence public perception.

My inspiration for building this project came from observing how false information spreads during global crises, especially during wars, geopolitical conflicts, cyber attacks, and political tensions. In modern conflicts, war is no longer fought only with weapons — it is also fought with information.

During conflicts between countries, social media often becomes a digital battlefield. Governments, intelligence agencies, propaganda networks, and coordinated groups may circulate misleading videos, manipulated media, or false narratives to influence public opinion or destabilize the opposing side. This type of strategy is often referred to as information warfare, where information itself becomes a tool used to shape narratives and perceptions. In many war scenarios, misleading or fabricated videos are shared online claiming:

Missile attacks that never happened

Military victories exaggerated or fabricated

Fake statements attributed to leaders

Old footage presented as new battlefield events

These videos can quickly go viral and reach millions of people before they are verified.

Such misinformation is often used to:

Create panic among civilians

Spread fear and uncertainty

Demoralize enemy populations

Manipulate international public opinion

Influence global media narratives

Seeing how easily people can be misled by viral videos during conflicts made me realize that verifying information manually is becoming extremely difficult.

What it does

Video Fact-Check is an AI-powered system that analyzes videos and verifies factual claims within them.

Instead of requiring users to manually verify every statement in a video, the system automatically:

Extracts the video transcript

Detects factual claims using AI

Searches for credible sources

Evaluates the claim using LLM reasoning

Generates evidence-backed fact-check results

Calculates an overall credibility score for the video

The goal is to make fact-checking faster, accessible, and scalable.

How we built it

he project is built as a full-stack web application with a modular architecture that allows future expansion.

Backend

The backend is built using FastAPI (Python) and handles the AI processing pipeline.

Core services include:

Transcription Service Extracts captions from videos or uses speech-to-text when captions are unavailable.

Claim Detection Service Uses an LLM to identify verifiable factual statements while ignoring opinions.

Search Service Retrieves reliable sources from the web using the Tavily API.

Fact Check Service Uses LLM reasoning to compare claims against evidence and determine their accuracy. Video URL ↓ Transcript Extraction ↓ Sentence Segmentation ↓ AI Claim Detection ↓ Web Search for Evidence ↓ LLM Fact Verification ↓ Credibility Score Generation

What we learned

Building this project helped me explore several important areas:

AI-assisted fact verification

LLM reasoning and prompt engineering

Full-stack development with FastAPI and Next.js

Designing scalable AI pipelines

Handling asynchronous workflows

Integrating web search with AI models

It also helped me better understand how misinformation spreads online and how difficult automated fact-checking can be.

Fact-checking is not only a technical challenge — it is also a societal challenge related to information trust and digital literacy.

What's next for VideoFactChecker

The long-term goal of this project is to help combat misinformation at scale.

Potential future features include:

Browser extension for real-time fact-checking on YouTube

Live claim highlighting while watching videos

Multi-language support

Database for storing previous analyses

Community-based fact verification

Integration with social media platforms

Batch analysis for journalists and researchers

Built With

• axios
• css
• docker
• fastapi
• gemini
• github
• gunicorn
• javascript
• next.js
• node.js
• tailwind
• tavilyapi
• typescript
• uvicorn
• youtubetransciptapi
• zustand