• 

  Landing Page

• 

  Our Search Results

TruthLens

🌍 Inspiration

Over 52% of U.S. adults get news from social media, where misinformation spreads rapidly (Pew Research). Political content today is shaped by polarization, AI-generated misinformation, and narrative framing.

Most tools try to declare what’s “true” or “false.”
TruthLens reveals narrative structure instead.

We quantify:

• Cross-source semantic agreement
• Narrative divergence
• Emotional intensity
• Source diversity

🧠 What It Does

Paste a headline or URL and TruthLens:

• Scrapes parallel coverage across publishers
  
• Generates 768-dimensional semantic embeddings
  
• Clusters articles using Elasticsearch kNN
  
• Computes a transparent confidence score
  
• Visualizes narrative groups and bias distribution
  

The dashboard includes:

- **Confidence Score (0–100)**
- **Narrative Divergence**
- **Bias Spectrum**
- **Emotional Framing**
- **Source Diversity**

⚙️ How We Built It

1️⃣ Bright Data

• SERP scraping
  
• Article HTML extraction
  
• Parallel coverage retrieval
  

2️⃣ Jina Embeddings (v3)

• 768-dimensional vectors
  
• Semantic representation of full article text
  

3️⃣ Elasticsearch (Elastic Cloud)

• dense_vector storage
  
• kNN vector search
  
• Similarity scoring
  
• Greedy clustering
  
• Metric aggregation
  

4️⃣ Deterministic Scoring

confidence =
(0.6 × agreement) +
(0.2 × diversity) +
(0.1 × (1 − emotional_intensity)) +
(0.1 × domain_age_factor)

Fully transparent. No opaque LLM reasoning.

🚧 Challenges

• Choosing the right vector similarity threshold
  
• Handling inconsistent article HTML structures
  
• Designing a neutral, defensible scoring formula
  
• Avoiding political labeling while preserving usefulness
  

---

🏆 Accomplishments

• Built a full ingest → embed → index → cluster → score → visualize pipeline
• Deep integration with Elasticsearch vector search
• Real-time parallel media coverage analysis
  
• Transparent, reproducible scoring system
  
• Production-ready dashboard UI
  

📚 What We Learned

• Vector search is powerful for narrative detection
• Agreement across independent sources is measurable
  
• Emotional intensity often correlates with divergence
  
• Deterministic AI builds trust
  

🚀 What’s Next

• 🌐 World heatmap of coverage origins
  
• 📈 Historical divergence tracking
  
• 🔔 Narrative shift detection
  
• 🧩 Chrome extension overlay
  

🔧 Full Technology List

Frontend

• Next.js (App Router)
  
• React
  
• Tailwind CSS
  
• Lucide Icons
  

Backend

• Next.js API Routes
  
• TypeScript
  

Search & Storage

• Elasticsearch (Elastic Cloud)
  
• dense_vector
  
• kNN search
  
• Vector similarity scoring
  
• Index mappings
  
• Google Cloud
  
• Aggregations
  

Embeddings

• Jina Embeddings v3 (768 dimensions)
  

Data Acquisition

• Bright Data API
  
• SERP API
  
• Web Scraper API
  

Analysis & Scoring

• Custom sentiment analysis logic
  
• Greedy threshold clustering algorithm
  
• Deterministic confidence scoring formula
  
• Elastic Cloud
  

Built With

• bright-data-api
• bright-data-serp-api
• bright-data-web-scraper-api
• custom-sentiment-analysis-logic
• deterministic-confidence-scoring-formula
• elasticsearch-(elastic-cloud)
• elasticsearch-aggregations
• elasticsearch-dense-vector
• elasticsearch-index-mappings
• elasticsearch-knn-vector-search
• environment-based-secret-management-(.env)
• greedy-threshold-clustering-algorithm
• jina-embeddings-v3-(768-dimensional-embeddings)
• lucide-icons
• mcp
• next.js
• next.js-api-routes
• node.js
• npm
• react
• tailwind-css
• typescript