VerifyPulse 🔍

Real-time news verification dashboard that scores breaking news by trustworthiness.

VerifyPulse aggregates news from multiple independent sources, clusters related stories, and calculates a confidence score based on how many credible outlets corroborate each claim. Instead of telling you what's true, it shows you the evidence — so you can decide.

Built as a solo founder prototype in 7 days. 100% free tech stack.

---

How It Works

  RSS Feeds (Reuters, AP, BBC, Al Jazeera, NDTV)
  + GDELT Global News Database
         │
         ▼
  ┌─────────────────┐
  │  DATA PIPELINE   │  Fetches every 15 min, deduplicates
  └────────┬────────┘
           ▼
  ┌─────────────────┐
  │  CLUSTERING      │  MiniLM embeddings + cosine similarity
  │  ENGINE          │  groups related articles into clusters
  └────────┬────────┘
           ▼
  ┌─────────────────┐
  │  CONFIDENCE      │  3-factor scoring: source count,
  │  SCORER          │  credibility, source diversity
  └────────┬────────┘
           ▼
  ┌─────────────────┐
  │  REST API        │  FastAPI with auto-generated docs
  │  (FastAPI)       │  at /docs
  └────────┬────────┘
           ▼
  ┌─────────────────┐
  │  DASHBOARD       │  Live feed with confidence badges,
  │  (HTML/JS)       │  filters, and "Why we trust this"
  └─────────────────┘

Confidence Scoring

Each story is scored on three factors:

Factor	Weight	What It Measures
Source Count	0-40 pts	How many independent sources report the story
Source Credibility	0-35 pts	Average credibility of reporting sources
Source Diversity	0-25 pts	Variety of source types (wire service, broadcaster, national)

The combined score maps to a confidence label:

Score	Label	Meaning
80-100	Verified	Multiple independent credible sources confirm
60-79	Likely Accurate	Strong sourcing, not fully independently confirmed
40-59	Developing	Some corroboration, story still evolving
20-39	Unverified	Limited sources, insufficient evidence
0-19	Disputed	Single source or contradictory reports

Tech Stack

Layer	Technology	Cost
Backend	Python 3.10+ + FastAPI	Free
Database	SQLite (WAL mode)	Free
NLP	sentence-transformers (MiniLM)	Free
Data Sources	RSS feeds + GDELT API	Free
Frontend	Vanilla HTML/CSS/JS	Free
Fonts	DM Sans + IBM Plex Mono	Free

Total cost: $0/month

Quick Start

Prerequisites

• Python 3.10+
• pip

Backend

cd backend
python -m venv venv
# Windows
venv\Scripts\activate
# macOS / Linux
source venv/bin/activate

pip install -r requirements.txt
python -m uvicorn app.main:app --reload

The API starts at http://localhost:8000 with auto-generated docs at http://localhost:8000/docs.

On first run, it downloads the MiniLM embedding model (~90 MB), creates a SQLite database, and begins fetching news automatically every 15 minutes.

Windows users: if you see a UnicodeEncodeError on startup, run with set PYTHONUTF8=1 before the uvicorn command, or just use start.bat which handles this automatically.

Windows shortcut

# From the project root, just double-click:
start.bat

This creates the venv and installs dependencies automatically on first run.

Frontend

# Just open the file in your browser
open frontend/index.html     # macOS
start frontend/index.html    # Windows

The dashboard connects to localhost:8000 and displays stories with confidence scores.

API Endpoints

Method	Endpoint	Description
GET	/api/stories	Clustered stories with confidence scores
GET	/api/stories/{id}	Story detail with full scoring breakdown
GET	/api/confidence-levels	Confidence level definitions
POST	/api/fetch	Manually trigger news fetch
GET	/api/articles	Raw articles with filters
GET	/api/sources	Source credibility table
GET	/api/stats	System overview

Full documentation: docs/API.md

Project Structure

verifypulse/
├── backend/
│   ├── app/
│   │   ├── main.py              # FastAPI app + lifecycle
│   │   ├── config.py            # Sources, regions, settings
│   │   ├── models/              # Pydantic data models
│   │   ├── routers/
│   │   │   ├── stories.py       # /api/stories endpoints
│   │   │   └── data.py          # /api/articles, sources, stats
│   │   └── services/
│   │       ├── rss_fetcher.py   # RSS feed parser
│   │       ├── gdelt_client.py  # GDELT API client
│   │       ├── database.py      # SQLite operations
│   │       ├── dedup.py         # Title similarity dedup
│   │       ├── clustering.py    # TF-IDF story clustering
│   │       ├── confidence.py    # 3-factor confidence scoring
│   │       └── scheduler.py     # Auto-fetch every 15 min
│   ├── tests/
│   │   └── test_clustering.py   # Unit tests
│   └── requirements.txt
├── frontend/
│   └── index.html               # Dashboard (single file, no build)
├── docs/
│   └── API.md                   # API documentation
├── Dockerfile
└── README.md

Future Roadmap (Phase 2)

• LLM-powered claim extraction (Claude API)
• Multilingual support (Hindi + more via XLM-RoBERTa)
• Semantic clustering with sentence-transformers
• Story timeline tracking (how stories evolve over time)
• Daily misinformation report generator
• Deepfake/image manipulation detection
• Browser extension for inline verification

Development Timeline

Day	What Was Built	Commits
1	Project setup, RSS parser, GDELT client	4
2	SQLite database, deduplication, auto-scheduler	4
3	TF-IDF clustering, confidence scoring, tests	4
4	API routers, documentation	3
5	Frontend dashboard with live feed	1
6	Source timeline, score ring, summary bar	2
7	README, Dockerfile, final polish	2

License

MIT

Author

Solo founder prototype — built to demonstrate that news verification can be automated, transparent, and free.