Disaster Relief ChatBot

Status: Ongoing | Funding: Howard University

Our project investigates the development of an accessible and resilient chatbot system to support communities during natural disasters and infrastructure disruptions. The central research questions motivating this work include: (1) How can large language models (LLMs) be adapted to provide concise, actionable information under conditions of limited connectivity? (2) What strategies ensure reliability and factual accuracy when delivering guidance about shelters, food, medical resources, and emergency alerts? (3) How can multimodal communication pathways—via internet-based mobile applications and SMS through Twilio—be integrated into a single system that maintains functionality during network outages?

Current approaches to disaster communication often rely on either internet-based platforms or broadcast alerts, which may not adequately serve vulnerable populations. Our work emphasizes dual-mode access: a mobile application that queries AI models when internet service is available, and an SMS-based interface that allows continuity of service when internet access is compromised.

Our present milestone focuses on implementing a prototype that integrates GPT-based conversational models with structured resource data (e.g., weather alerts, local emergency management updates) while incorporating safeguards against hallucinations and unverified outputs. Future stages will evaluate system performance in simulated disaster scenarios and explore user-centered design in collaboration with social scientists and behavioral researchers.

Ultimately, this research aims to demonstrate how AI-driven conversational agents can contribute to more resilient disaster preparedness and response infrastructures, particularly in contexts where access to information is most fragile.

Contributors

Progress Timeline

• 2025-Present