Deadzone

During floods, cyclones, and other disasters, internet connectivity often becomes unreliable, making it difficult for people to communicate their safety or request help. Families, NGOs, journalists, and emergency responders lose visibility into affected areas, while regions with no communication become "dead zones," delaying rescue and aid. This communication gap is especially critical in Bangladesh, where disasters frequently disrupt conventional communication channels.

DeadZone is a disaster communication platform that enables citizens to report their safety ("I'm Alive") or request emergency assistance in Bangla or English through a Telegram bot. It uses Bangla natural language processing (NLP) to interpret free-text messages, identify the type of report, extract locations, categorize needs, and assign priority based on urgency. The processed information is stored and displayed on a live coordinator dashboard featuring an H3-powered Dead Zone Heatmap, real-time pulse locations, and prioritized aid requests. By revealing both active reports and areas with little or no communication, DeadZone helps NGOs, volunteers, journalists, and emergency responders coordinate relief efforts, allocate resources efficiently, and reach affected communities faster during disasters.

Tech Stack and Third-Party Components Languages Python 3.12 JavaScript HTML5 CSS3 Backend FastAPI Uvicorn Pydantic Database & Storage Supabase PostgreSQL Geospatial Uber H3 h3-js Frontend Leaflet.js APIs & Integrations Telegram Bot API Supabase API Deployment Render GitHub Geospatial Uber H3 – Hexagonal hierarchical geospatial indexing for aggregating reports and generating the Dead Zone Heatmap. h3-js – JavaScript library for rendering H3 hexagons on the interactive map.

ChatGPT (OpenAI): Assisted with brainstorming, documentation, code refinement, debugging, and UI/content generation. Rule-based Bangla NLP: Used to parse free-text emergency messages, classify needs, detect urgency, and extract locations from user reports. (Custom implementation; no pre-trained language model is used in the current MVP.)