How many times have you wondered if your Hackathon project was good enough or if your idea was worth a prize? I know right? That's why our team created Devscale, a platform that gives insights and statistics about your project's likelihood to win a Hackathon. It's as simple as that.
What it does
Bear with me: you really would like to know how your project and idea compares to other Hackathon winners and thus find out what is the likelihood of your project to win. Devscale lets you do exactly that. You access our website, input your headline and submission description, and we will provide meticulous and actionable statistics and insights about your project and the likelihood for it to win. For us to achieve this feat, we created a machine learning model trained with over 15,000 hackathon projects. For example, our Devscale model will tell you that you might have a 60% chance of winning (and thus 40% chance of losing) based on your current project description.
How we built it
Front-ent: We used carefully crafted HTML, JSX, CSS, and React for the front-end. All icons are from Font Awesome and any images, logos, and SVGs were gently handmaid from scratch by our team.
Dataset: Our dataset was built using UiPath, a robotic process automation tool that facilitates the process of scraping websites. We were able to scrape 15000 projects and details from Devpost. The Dataset contains texts and labels defining if the project is a winner or not. The file was uploaded to a Google Cloud Storage Bucket for further manipulations.
Google Cloud AutoML Model: Our Machine Learning model was trained and deployed on Google Cloud. We selected the Natural Language Processing Text Classification feature to make predictions on the model.
Challenges we ran into
Setting up a REST API in a rush during the last hours of the competition was definitely a challenge we ran into. _Our team is currently in and out of programming competitions this weekend besides studying for college exams _; therefore tying up the knots and connecting all the cables between front-end, back-end, and datasets under pressure is absolutely never-racking. In addition to that, the automatic scraping was not really smooth due to some projects having symbols, different languages, and emojis in their descriptions. We had to process the data before training the model.
Accomplishments that we're proud of
Bringing our idea to life to be the first initiative to make Hackathon projects more likely to win and allowing users to get more insights and statistics into their project submissions is absolutely amazing. Also, conquering the challenges cited above is absolutely something our team is absolutely proud of.
What we learned
Our team members complete themselves based on their skills; therefore, we learned a lot. Nathan, our team's Google Cloud guru learned about how to connect the cables between React components and Rest APIs based on Node.js. Mauricio, our team's React monk, also learned how to write API calls from scratch directly from React components.
What's next for Devscale
We would like to provide even more actionable statistics for hackers to rejoice. To do that, we are on a mission to improve our Devscale model's accuracy by training it with different machine learning algorithms and increasing our dataset of winning projects.