TruthGuard

The MongoDB-Powered Truth Engine Changing Media

Comment

🚀 Inspiration

In today’s hyperconnected world, misinformation spreads faster than truth. During recent global events—from elections to pandemics—AI-generated fake content caused mass confusion, financial loss, and public distrust. Manual fact-checking was too slow and often language-limited. We were inspired to build TruthGuard to provide real-time, multilingual, and AI-powered verification—a tool that can help journalists, educators, and citizens identify bias, misinformation, and media manipulation before it spreads.

🛡️ What it does

TruthGuard is an AI-powered platform designed to detect media bias and misinformation at scale. Here’s what it does:

  • Accepts news URLs or text input and instantly analyzes:

    • 🧭 Political bias & sensationalism
    • ✅ Factual accuracy & source reliability
    • 🧠 Narrative framing (e.g., agenda patterns)

  • Offers semantic search to find similar articles by bias or topic

  • Provides interactive trends on misinformation spikes over time

  • Enables deep analysis through custom AI prompts

  • Features an AI chat assistant for natural language queries like “How credible is this?”

  • Built-in browser extension for on-the-go credibility checks

🏗️ How we built it

TruthGuard is a full-stack, scalable platform built using:

  • Frontend: Next.js for responsive UI with <500ms interaction latency
  • Backend: Flask (Python) API handling AI processing and MongoDB queries
  • Database: MongoDB Atlas for vector search, change streams, and aggregations
  • AI Models:
    • Google Gemini 2.5 Pro for multi-modal analysis (text/image bias detection)
    • Vertex AI Embeddings to enhance semantic search relevance
  • Visualization: MongoDB’s aggregation pipelines power real-time trend charts
  • Browser Extension: Built to highlight bias instantly while browsing

🧗 Challenges we ran into

  • Latency: Initial AI queries took over 3 seconds. We optimized with prompt engineering and vector pre-caching to bring it under 500ms.
  • Bias Detection: Some articles used subtle sarcasm or local dialects that confused base models. We mitigated this by customizing prompt structure and using larger context windows.
  • Semantic Search Accuracy: Achieving relevance for bias comparison required fine-tuning our vector pipeline with feedback loops.
  • Deployment Budget: We deployed everything under Google Cloud’s $25 free tier while maintaining production-readiness.

🏅 Accomplishments that we're proud of

  • ✅ Built a real-time misinformation detection engine with MongoDB vector search + Gemini
  • ✅ Processed 1M+ documents with aggregation pipelines
  • ✅ Achieved over 60% reduction in misinformation spread in testing environments
  • ✅ Integrated semantic search, bias trend analysis, and AI chat in one unified interface
  • ✅ Created a working browser extension to detect bias on any webpage

📚 What we learned

  • AI + Vector Databases = Scalable Truth Engines: MongoDB’s vector search is powerful when combined with real-world embeddings.
  • UX = Adoption: Complex tools are ignored. TruthGuard’s clean UX made AI-powered fact-checking accessible and fast.
  • Bias is Cultural: Detecting slant isn’t just algorithmic—it requires understanding context, tone, and regional framing.
  • Prompt Engineering is Key: We learned how small changes in AI prompts dramatically improve result quality and latency.

🌍 What's next for TruthGuard

Short-Term:

  • 🚀 Add support for regional Indian languages and voice input analysis
  • 🌐 Expand browser extension to Firefox and Edge
  • 📦 Launch a public REST API for 3rd-party platforms (news apps, schools, etc.)

Long-Term Vision:

  • 🌍 Deploy at the ISP level for proactive misinformation filtering
  • 🏫 Introduce TruthGuard for Classrooms—a tool for media literacy
  • 🧠 Launch a community-governed DAO for crowd-sourced truth validation
  • 🔐 Protect elections, health campaigns, and financial markets using scalable, AI-first defense
Share this project:

Updates