Crisis Map AI

CrisisMap AI: Harnessing the power of AI and real-time data to monitor and detect global crises, providing timely insights and responses to humanitarian challenges.

Comment

Inspiration

In today's world of frequent natural disasters and humanitarian crises, finding relevant information quickly can save lives. I built CrisisMap AI to solve the challenge of information overload during critical situations, inspired by events like California wildfires and recent earthquakes where fragmented data hampered effective response.

What It Does

  • Ingests crisis data from multiple sources (official databases, news feeds, social media)
  • Converts crisis reports into vector embeddings for powerful semantic search
  • Processes natural language queries about disasters and humanitarian crises
  • Provides AI-generated summaries and analysis of complex situations
  • Supplements database information with real-time web scraping for current data
  • Enables unified access to previously fragmented crisis information

How I Built It The architecture combines powerful technologies:

  1. Data Layer: MongoDB Atlas with vector search for semantic querying
  2. AI Processing: Microsoft's Phi-3-mini-4k-instruct model for language understanding
  3. Data Pipeline: Processing various datasets (WHO, EM-DAT, USGS) with AI-enrichment
  4. Web Scraping: Using Beautiful Soup to augment results with latest information
  5. API Layer: FastAPI providing a performant interface for queries

Challenges & Learnings

Building this system taught me about:

  • Data Standardization: Unifying diverse crisis data formats
  • LLM Integration: Optimizing the Phi-3 model within resource constraints
  • Vector Search: Fine-tuning MongoDB's vector search parameters
  • Security: Implementing proper credential protection

What's Next I plan to expand with real-time social media monitoring for early detection, multi-language support for global accessibility, advanced crisis prediction using historical patterns, and interactive mapping and visualization. By continuing development, I hope to enable more effective and informed responses to global crises.

Built With

Share this project:

Updates