TrueStream – AI Misinformation Guardian

AI-powered real-time misinformation detection for images, videos, and URLs. Verify before you share with evidence-based scoring, reverse search, and multimodal analysis.

Comment

Inspiration

Misinformation spreads faster than truth.
A single misleading video can reach millions before anyone verifies it.

We kept asking ourselves a simple question:

Why is it easier to share content than to verify it?

In the age of deepfakes, edited clips, and recycled images from years ago, people need a simple tool that says:

"Wait. Let’s check this first."

TrueStream was born from that idea — an AI-powered guardian that verifies content before it spreads.

What it does

TrueStream is a real-time misinformation detection platform.

Users can upload:

  • An image
  • A short video
  • Or paste a URL

Within seconds, the system:

  1. Extracts metadata and media frames
  2. Performs reverse image search
  3. Checks authoritative news sources
  4. Transcribes and analyzes audio (for videos)
  5. Uses multimodal AI reasoning

It then produces:

  • A credibility score (0–100)
  • A clear verdict (Likely True / Inconclusive / Likely Misleading)
  • Transparent evidence
  • A recommended safe caption

It doesn’t just give a score — it explains why.

How we built it

TrueStream is built as a Progressive Web App (PWA) using:

  • Next.js (full-stack architecture)
  • TypeScript
  • TailwindCSS
  • IndexedDB for offline history
  • Service Worker for offline capability
  • Google Gemini API for AI reasoning
  • Google Custom Search API for reverse image lookup
  • NewsAPI for news verification

The system works in two modes:

  • Mock Mode (default) → deterministic results for reliable demo without API keys
  • Live Mode → full real-world API integration

The architecture follows a structured pipeline:


Input → Media Analysis → Evidence Retrieval → AI Reasoning → Scoring → Transparent Output

Everything is modular, scalable, and production-ready.

Challenges we ran into

One of our biggest challenges was avoiding a “black-box AI” experience.

We didn’t want the system to simply say “Fake” without context.

So we designed a hybrid approach:

  • Deterministic evidence scoring
  • Signal weighting
  • AI-generated explanation

Another challenge was implementing true offline-first behavior while still supporting live API integrations.

Ensuring proper PWA installation, caching behavior, and cross-browser compatibility required careful debugging and optimization.

Accomplishments that we're proud of

  • Fully installable PWA that works offline
  • Deterministic mock mode for stable demos
  • Transparent evidence timeline instead of vague AI judgment
  • Bilingual interface (English + Arabic)
  • Clean UI with animated credibility gauge
  • Modular architecture ready for newsroom integration

Most importantly — we built something usable, not just conceptual.

What we learned

We learned that misinformation detection isn’t just about AI — it’s about trust.

Users don’t trust scores.
They trust explanations.

We also learned how to combine rule-based signals with LLM reasoning to produce explainable, human-readable outputs.

Technically, we deepened our understanding of:

  • Progressive Web App architecture
  • Multimodal processing pipelines
  • API fallback strategies
  • Frontend performance optimization

What's next for TrueStream – AI Misinformation Guardian

  • Browser extension for real-time verification before posting
  • API access for media organizations
  • Expanded multilingual support
  • Fine-tuned detection models for regional misinformation
  • Native mobile wrapper

Our long-term goal is simple:

Make verification as easy as sharing.

Built With

  • googlecustomsearchapi
  • googlegeminiapi
  • indexeddb
  • newsapi
  • next.js
  • react
  • serviceworker(pwa)
  • tailwindcss
  • typescript
  • vercel
Share this project:

Updates