Rate a Hack!

The hardest part of a hackathon is finding a truly original idea. We use our algorithm to rate your idea against thousands of past projects and measure its originality in seconds.

Comment

Inspiration

One of the hardest parts of a hackathon isn't the building of an app, it's knowing whether your idea is actually original.

What it does

We built a machine-learning tool that compares a team's hackathon idea against over 65,000 past hackathon projects posted to Github and Devpost. The algorithm develops a similarity score by comparing key phrases and words in the user generated to the clustering made through the algorithms.

How we built it

We used the official Github API and Devpost data to create a JSONL database of past projects. SemanticTransformer, FAISS, and a house-made algorithm was utilized to produce an originality score for a project idea. When the user submits an idea, a JSON packet is sent to the backend which compares the submission to the multidimensional array of the training data. From there, a similarity score is generated and sent back to the front end in another JSON packet. The similarity score was displayed and a suggestion algorithm was utilized to suggest feedback on particular areas the project can focus on as well as improvements.

Challenges we ran into

The main challenges of our project were the development of the algorithm, specifically calibrating the weights of key words and ensuring it scores based on topic originality rather than semantics. Collecting data also took a fair amount of time due to rate-limits and staying within the terms and conditions of Github and Devpost.

Accomplishments that we're proud of

We are proud of the hand drawn images and animations that occurred, making a fully-functional algorithm which clusters projects together based on similarities, our calibration of the algorithm such that the scores were distributed across a large range of percentages, creating a database full of previous hackathon submissions, and our front-end with intuitive UI.

What we learned

How to utilize machine-learning and algorithms on a large set of data to ensure it produces efficient and accurate results.

What's next for Rate a Hack?

In the future, we hope to make this a fully functional webpage with a dedicated server and backend rather than running it on a local machine. We also hope to expand the training data to include over 500,000 past hackathon projects (all of Devpost).

Built With

Share this project:

Updates