911 receives too many calls and is far to understaffed. Some of these calls aren't urgent and can be routed to non-emergency services. Some of them can be directly routed to the emergency response team they might need. We systematically handle this routing through one single number.
What it does
You text our number, and our chatbot understands your needs and patches you to the correct services. We not only connect you to 911, but other disaster relief agencies such as American red cross or the blue cross society. We also predict which neighborhoods will be affected the most, so resources can be distributed according to need.
How we built it
We used IBM's NLP to understand a user's intent when they text us, based on which we connect them to the right people. We also scrape tweets and use IBM's studio to build custom models to cluster these tweets into neighborhoods, so we know which neighborhood needs the most help. We also analyze tweets and pick out actionable items, so a volunteer can work on them.
Challenges we ran into
Since we were just a team of two, this was a lot of work for just 24hrs. Finding datasets for our Machine Learning systems was the hardest since most emergency services keep their data private. We were both new web dev, so building an entire web app was a slight bump in the road.
Accomplishments that we're proud of
Actually implementing the core machine learning parts of the hack as well developing a full functioning UI for the demo.
What we learned
What's next for Wats (going) on
- Include translations for volunteers who may not speak the victim's language.
- Include GeoSpatial data into our machine learning models, to make them even more accurate.