SafeGraph AI

Inspiration

During disasters such as floods, heatwaves, cyclones, and earthquakes, people often struggle to find reliable, understandable, and situation-specific information. Existing resources are usually scattered across different websites and applications, making it difficult for students and communities to understand what actions they should take before, during, and after a disaster.

This inspired us to build SafeGraph AI, an intelligent disaster preparedness and emergency response education platform that combines Knowledge Graphs, Agentic AI, and explainable AI to make disaster-related knowledge easier to learn, understand, and apply.

What We Built

SafeGraph AI provides disaster-specific learning courses, quizzes, progress tracking, emergency guidance, and an AI-powered Ask Doubts feature.

The core of the system is a Knowledge Graph built using Neo4j, which connects disasters, causes, risks, symptoms, preparedness measures, emergency actions, and safety guidelines. This structured knowledge is combined with an agentic AI pipeline that retrieves relevant information, reasons over the knowledge, and generates contextual and explainable responses.

The platform also includes an administrative module for managing learning content, monitoring student progress, analyzing quiz performance, and managing disaster-related information.

We also integrated a heatwave prediction module that provides weather-based risk insights and preparedness guidance, helping users understand potential heatwave conditions and take preventive action.

How We Built It

Frontend: Next.js, React, TypeScript
Backend: Node.js, Express.js
Database: NeonDB
Knowledge Graph: Neo4j
AI Service: Python, FastAPI
AI/LLM: Gemini API with agentic retrieval and reasoning
Authentication: JWT and Google OAuth
Deployment: Vercel and Render
Weather & Location Data: Open-Meteo, NASA POWER, OpenStreetMap/Overpass API

The system is designed as a modular architecture where the web application communicates with the backend, database, Knowledge Graph, and dedicated AI service.

What We Learned

Through this project, we learned how to integrate Knowledge Graphs with Large Language Models instead of relying only on direct LLM responses. We also gained practical experience in agentic AI workflows, retrieval and reasoning, API integration, authentication, database management, cloud deployment, and building responsive web applications.

One of our biggest learnings was that an AI system should not only provide an answer but should also provide context and reasoning that users can understand and trust.

Challenges We Faced

One of the major challenges was connecting unstructured disaster-related information with a structured Knowledge Graph while keeping the generated responses relevant.

We also faced challenges in integrating the AI service with the main application, handling authentication across different services, deploying multiple components, and maintaining reliable communication between the frontend, backend, database, Knowledge Graph, and AI microservice.

Another challenge was designing the system so that AI-generated responses remain grounded in relevant disaster knowledge rather than producing generic answers.

Future Scope

SafeGraph AI can be extended with real-time disaster alerts, more disaster-specific prediction models, multilingual support, personalized emergency recommendations, live location-based risk analysis, and integration with government and emergency-response data sources.

Our goal is to evolve SafeGraph AI into an intelligent platform that helps people learn, understand, prepare, and respond to disasters more effectively.