A safe path from Sather Gate to Whole Foods in Berkeley.
The same safe path with crime frequency visualizations turned off.
The fastest, but least safe path from Sather Gate to Whole Foods in Berkeley.

Inspiration

In the past few years, we have witnessed a large spike in crime in the US, making the streets less safe and putting innocent people at risk. HomeSafe is an attempt to solve this problem by providing a simple solution to help people navigate the streets without taking unnecessary risks.

What HomeSafe does

HomeSafe is a cross-platform web application that aggregates crime data from multiple different sources - including real-time radio data analyzed using voice recognition and NLP - to assess the safety of different areas and help users navigate around areas deemed particularly unsafe, while still allowing them to comfortably reach their destination.

How we built it

Frontend: HTML/CSS/JavaScript
Backend: Flask
Database: Cockroach DB Serverless
Voice Recognition: Google Cloud Services
NLP: Cohere

We aggregate data from multiple different sources, including crime data scraped from the Berkeley PD and real-time reports from police scanners. For this, we use Google Cloud Services to transcribe the incoming audio and then extract locations of interest using a custom Cohere model. All of the extracted data is stored using Cockroach DB.

After extracting the location and severity of crimes based on our data, we assign safety ratings to areas based on crime frequency, severity and population density. We then use these ratings to assign weights to every path taking into account distance, safety and user priority. Finally, we find the shortest path using the GraphHopper routing engine with custom weights.

Challenges

This was our first foray into voice recognition and NLP. Thus, the initial learning curve was quite steep, as we had to learn many new concepts and technologies.