Team Name: PerlT Members: Claire Chen (ccz) Robert Chen (robertch) Parth Shastri (pdshastr) Charlotte Wang (crw2)

Inspiration

Long waits at Au Bon Pain, and safety during the COVID-19 pandemic.

What it does

It is a mobile app that provides a real-time heatmap of dining locations, as well as predictions of future crowdedness to help you better plan your dining choices at CMU.

How we built it

GEOVID uses GPS services to automatically map out the student population. GEOVID does this without compromising on privacy, anonymously uploading information to our cloud backend.

After testing our application with a walk-around campus, we’ve developed a three-step error correction process to reduce noise in the location data and improve the detection of population density.

We then constructed heat maps of historical and real-time data with JavaScript, using generated paths based on our own location information to simulate user data.

We also performed statistical analysis on our generated location dataset. After testing different hyperparameters, we settled on a GRU with 16 hidden units and we plan to integrate this model into our Flask API to predict future dining trends.

We’ve designed a website where you can play with these demos yourself. Our vision is to mitigate the risks of COVID transmission by mapping population hotspots on CMU’s campus and beyond.

Challenges we ran into

Properly containerizing our backend service and deploying it to our production server was also annoying. We burnt an hour trying to wrangle a UWSGI Dockerfile for proper static file serving.

We also needed data to test our app with, so we had to manually generate data by walking around campus.

Accomplishments that we're proud of

We wrote our own python backend! Some of the data wrangling with the coordinates were also quite fun.

What's next for GEOVID @ CMU

Integrate the ML model into the backend, and finish building out our IOS application from the wireframes.

Share this project:

Updates