🌐 Global News Intelligence Platform

Global news analytics with GDELT + AI + modern data stack

Live Demo • Features • Architecture • Tech Stack • Quick Start • Cost Efficiency

🎯 Overview

A full-stack data engineering project that ingests, processes, and visualizes 100,000+ daily global news events from the GDELT Project. Includes AI chat for natural language queries and a live analytics dashboard.

📊 By the Numbers

What is GDELT?

The GDELT Project monitors the world's news media from nearly every country in 100+ languages, identifying people, locations, themes, and emotions driving global society.

📸 Dashboard Preview

Home - KPIs & Trending News

Emotions - GKG Mood Analysis (NEW!)

Analytics - Actors & Countries

AI Chat - Natural Language Queries

RAG Chat - AI Analysis of World Events

Feed - Event Stream

✨ Features

🏗️ Architecture

┌─────────────────────────────────────────────────────────────────────────┐
│                         PRODUCTION PIPELINE ARCHITECTURE                 │
└─────────────────────────────────────────────────────────────────────────┘

              ┌──────────────┐          ┌──────────────┐
              │ GDELT Events │          │  GDELT GKG   │
              └──────┬───────┘          └──────┬───────┘
                     │                         │
                     └────────────┬────────────┘
                                  ▼
┌─────────────────────────────────────────────────────────────────────────┐
│  INGESTION (Every 15 min via cron-job.org → workflow_dispatch)                                                │
│  GitHub Actions → Dagster → Polars (10x faster) → custom validation      │
└─────────────────────────────────────────────────────────────────────────┘
                                  │
                                  ▼
┌─────────────────────────────────────────────────────────────────────────┐
│  TRANSFORMATION                                                          │
│  dbt Core: staging (stg_events) → marts (fct_daily, dim_actors, etc.)   │
└─────────────────────────────────────────────────────────────────────────┘
                                  │
                                  ▼
┌─────────────────────────────────────────────────────────────────────────┐
│  STORAGE & AI                                                            │
│  MotherDuck (DWH) ← Voyage AI (Embeddings) → Cerebras LLM (RAG/SQL)     │
│  └── gkg_emotions: Fear, Joy, Tone, Topics                              │
└─────────────────────────────────────────────────────────────────────────┘
                                  │
                                  ▼
┌─────────────────────────────────────────────────────────────────────────┐
│  PRESENTATION                                                            │
│  Streamlit: HOME | FEED | EMOTIONS | AI Chat | ABOUT                    │
└─────────────────────────────────────────────────────────────────────────┘

Data Flow (ELT Pipeline)

1. Extract: GDELT Events API + GKG Feed → Polars (10x faster than Pandas)
2. Validate: Custom schema + threshold data quality checks
3. Load: Deduplicated data into MotherDuck (serverless DuckDB)
4. Transform: dbt models create staging views and mart tables
5. Emotions: GKG data → Extract tone, fear, joy, topics (rolling 24h)
6. Embed: Voyage AI generates vectors every 12 hours
7. Serve: Streamlit dashboard with AI chat (SQL + RAG modes)

🛠️ Tech Stack

Data Engineering

AI/ML

Frontend

Skills Demonstrated

• Python (Polars, Pandas, RegEx, API integration)
• SQL (Complex queries, window functions, dbt models)
• Data Quality (custom data quality validation, schema testing)
• ELT Pipelines (Extract, Load, Transform with dbt)
• CI/CD (GitHub Actions, cron scheduling)
• Vector Search (Embeddings, cosine similarity, RAG)

🚀 Quick Start

Prerequisites

• Python 3.10+
• MotherDuck Account (free tier)
• Cerebras API Key (free tier)

Installation

# Clone the repository
git clone https://github.com/Mohith-akash/Global-News-Intel-Platform.git
cd Global-News-Intel-Platform

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Linux/Mac
# or
.\venv\Scripts\activate   # Windows

# Install dependencies
pip install -r requirements.txt

Configuration

Create a .env file in the project root:

MOTHERDUCK_TOKEN=your_motherduck_token
CEREBRAS_API_KEY=your_cerebras_api_key
VOYAGE_API_KEY=your_voyage_api_key  # Optional: enables RAG mode

Run the Dashboard

streamlit run app.py

Run the Pipeline Manually

# Polars-powered ingestion (15-min schedule)
python -m dagster job execute -f etl/pipeline_polars.py -j gdelt_ingestion_job

# Embedding generation (12-hour schedule)
python -m dagster job execute -f etl/embedding_job.py -j gdelt_embedding_job

# Run dbt models
cd dbt && dbt run

💰 Enterprise Tools vs My Stack

This project demonstrates how to achieve enterprise-grade capabilities at zero cost:

Key Insight: MotherDuck's native vector search eliminates the need for a separate vector database like Pinecone.

🔄 Technology Evolution

This project evolved through multiple iterations to optimize for cost and performance:

Data Warehouse

❄️ Snowflake (trial) → 🦆 MotherDuck (free tier)

• Started with Snowflake trial for learning enterprise DWH
• Migrated to MotherDuck to eliminate costs while keeping SQL compatibility

AI/LLM Provider

✨ Gemini 2.0/2.5 Flash → ⚡ Groq (Llama 3.3 70B) → 🧠 Cerebras (Llama 3.1 8B → GPT-OSS 120B)

• Tested Gemini models for natural language queries
• Tried Groq's fast inference with larger Llama models
• Settled on Cerebras for reliable free tier and good performance
• Moved to GPT-OSS 120B after Cerebras archived the Llama 3.1 models

RAG Embeddings

🚀 Voyage AI (embeddings) + 🦆 MotherDuck (vector search)

• Voyage AI creates 1024-dim embeddings for semantic search
• MotherDuck's native array_cosine_similarity() replaces Pinecone
• Dual-mode AI: SQL for precise queries, RAG for semantic exploration

Key Learning: The best tool isn't always the most expensive—it's the one that solves your problem within constraints.

📁 Project Structure

gdelt_project/
├── app.py                    # Streamlit dashboard entry point
├── src/                      # Core modules
│   ├── config.py             # Configuration constants
│   ├── database.py           # Database connection
│   ├── queries.py            # SQL query functions
│   ├── ai_engine.py          # LLM/AI setup (Cerebras + LlamaIndex)
│   ├── rag_engine.py         # RAG engine (Voyage AI + vector search)
│   ├── data_processing.py    # Headline extraction
│   ├── utils.py              # Utility functions
│   └── styles.py             # CSS styling
├── etl/                      # Data pipeline
│   ├── pipeline_polars.py    # 🆕 Polars ingestion + custom validation
│   └── embedding_job.py      # 🆕 12-hour embedding generation
├── dbt/                      # 🆕 dbt transformation layer
│   ├── dbt_project.yml       # dbt configuration
│   ├── profiles.yml          # MotherDuck connection
│   └── models/
│       ├── staging/          # stg_events (cleaned data)
│       └── marts/            # fct_daily_events, dim_actors, dim_countries
├── components/               # UI components
│   ├── render.py             # Dashboard rendering
│   ├── ai_chat.py            # AI chat interface
│   ├── emotions.py           # GKG emotions tab
│   └── about.py              # About page
├── requirements.txt          # Python dependencies
├── .env                      # Environment variables (not in repo)
└── .github/workflows/
    ├── gdelt_ingest.yml          # 15-min Polars ingestion
    ├── gdelt_embeddings_12hr.yml # 12-hour embedding job
    └── health_monitor.yml        # uptime checks + ntfy alerts

🔮 Future Enhancements

• Add dbt transformations for advanced modeling ✅ Done!
• Upgrade to Polars for faster processing ✅ Done!
• Add data quality validation ✅ Done!
• Implement event clustering with ML
• Add email/Slack alerts for crisis events
• Expand AI chat with multi-turn conversations
• Add export functionality (CSV, PDF reports)

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📬 Contact

Mohith Akash

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

_{Built with ☕ and curiosity • Data sourced from GDELT Project}