You never know which will come first, tomorrow or the disaster. Once a disaster occurs, beyond the 72 golden hours, the death rate increases dramatically.

However, how to race against time to better perform supplies estimation and job assigning?

Big Data and AI can help us accurately predict the amount of aid and supplies needed for victims, leading to better allocation of supplies and manpower by government agencies and relief organizations.

The database (credit appen) for this project is from the real world: 26,216 messages, 3 genres (direct, news, and social), 36 categories.

The project was built on a deep learning neural network model, with 95% average Recall Scores for all of the categories.

Why Recall Score? It more focuses on quantity, whether or not irrelevant messages are also included. e.g., "I have food, I can help my neighborhood." maybe treat as an SOS message. However, SOS is so important that we should take the risk of classifying irrelevant as SOS.

• ETL Pipeline: Automatic ETL pipeline for data processing and database dumping.
• ML Pipeline: Featuring engineering, model training, and model deployed on production.
• Web Development: REST API development for model inference. Fully responsive with well design.
• DevOps: Containerized by Docker. Deployed on AWS ECS with least computing resource and management required.
• Security & Resilience: Well-architected framework with least privilege, access control, and auto scaling.