Skip to content

README.md

Latest commit

 

History

History
109 lines (81 loc) · 3.84 KB

README.md

File metadata and controls

109 lines (81 loc) · 3.84 KB

🚨 Disaster Alert System – HackNova

Team Name: HackNova
Team Members:

  • Anish Bandal (Team Lead)
  • Priti Chavan
  • Swati Mane
  • Atharva Gitaye

📌 About the Project

The Disaster Alert System is an AI-powered real-time monitoring dashboard that detects, verifies, and tracks natural disasters from social media (Reddit).
It provides emergency teams with credible alerts, severity levels, and geolocation mapping for better disaster management.

🔑 Key Features

  • 🌐 Hybrid AI Pipeline

    • Custom ML models for sarcasm detection, binary disaster classification, and disaster type classification.
    • Gemini API fallback to re-check low-confidence predictions.

  • 📍 Location Extraction & Mapping

    • NER + OpenStreetMap geocoding to pinpoint disaster locations.

  • 🔥 Credibility & Severity Scoring

    • Multiple user posts increase credibility.
    • Auto-assigns severity levels (Low, Medium, High, Critical).

  • 📊 Next.js Frontend Dashboard

    • Real-time map visualization & incident logs from Firebase Firestore.

  • ⚡ Reddit Integration

    • Streams posts from disaster-related subreddits in real time.

🤖 AI Models & Pipeline

Our AI system uses a multi-stage pipeline for accurate disaster detection:

  1. Sarcasm Detection Model

    • Filters out sarcastic or ironic posts.
    • Example: "Wow, what an amazing earthquake 🙄" → Marked as sarcasm → Ignored.

  2. Binary Disaster Classifier

    • Determines if a post actually describes a disaster or not.
    • Example: "Severe flooding in Pakistan displaces thousands" → Disaster: True.

  3. Multiclass Disaster Classifier

    • Identifies which disaster it belongs to (Earthquake, Flood, Fire, Hurricane, etc. – 11 types).

  4. Location Extraction (NER)

    • Uses Named Entity Recognition (NER) to detect locations in text.
    • Integrated with OpenStreetMap for latitude/longitude mapping.

🔄 Hybrid Structure (Models + Gemini)

  • If the confidence score from our custom ML models is high, we trust the prediction.
  • If confidence is low or uncertain, the text is re-verified using Google Gemini API.
  • The most reliable output is then stored in Firestore.
  • This ensures high accuracy even when custom models are unsure.

📊 System Architecture

flowchart TD
    R[Reddit API / Social Media] -->|New Post| B[Flask Backend]
    B -->|Sarcasm Model| S[Sarcasm Detection]
    S -->|Binary Classifier| D[Binary Disaster Classifier]
    D -->|Multiclass Disaster Model| M[Multiclass Disaster Model]
    M -->|NER + Geocoding| L[Location Extraction]
    L -->|Gemini Verification| G[Gemini Disaster Type Check]
    G -->|Confirmed Disaster| F[Firebase Firestore]
    F -->|Recent Incidents API| FE[Next.JS Frontend Dashboard]
Loading

🛠️ Tech Stack

  • Frontend: Next.js + Tailwind CSS
  • Backend: Flask (Python)
  • AI Models: Custom Transformers + Gemini API (hybrid)
  • Database: Firebase Firestore
  • Data Source: Reddit API (PRAW)

🚀 How It Works

  1. Reddit posts are streamed and passed to the backend.
  2. Posts go through AI pipeline → Sarcasm filter → Binary classifier → Disaster type classifier.
  3. Low-confidence results checked with Gemini.
  4. Locations extracted and geocoded.
  5. Events stored in Firestore with:
    • disaster_type, severity, credibility_score, reports_count, source_link, resolved.
  6. Frontend dashboard fetches incidents and displays them on a map and analytics dashboard.

📂 Project Status

✅ Frontend with real-time Firestore sync
✅ Backend with hybrid AI pipeline
✅ Reddit ingestion & Firestore storage
✅ Severity & credibility scoring
🚀 Fully integrated Emergency Dashboard