What it does
You can give us your Kickstarter project idea and we'll use - oooh! - machine learning to tell you if it will be successful or not.
Fun fact: Did you know that machine learning is always right and can be completely trusted to make decisions that make lasting impacts in our communities? (ɔ◔‿◔)ɔ ♥
How we built it
The frontend is Vue! The backend is Flask! The machine learning is BERT! The database is Cockroach! The hosting is Google! Zendaya is Meechee!
Our data comes from Kaggle, Mickaël Mouillé's excellent dataset of over 300,000 projects that we loaded into CockroachDB.
Challenges we ran into
While we were able to set up BERT to perform semantic similarity calculations on our dataset, we discovered that our simple Google App Engine setup was severely limited in how much RAM it could use, meaning we could not manage to actually run these calculations in the cloud without a complete re-architecting of our application...
The live demo uses simulated data. But it works! We promise! See
bert.py in our Github for proof!
Accomplishments that we're proud of
Despite a busy semester and many miles of distance, we were able to build a project together!
What we learned
Teams work, databases are tricky, and machine-learning is resource heavy!