InsideRadar | Devpost

Home page!!
Light mode!!
Home Page 2
Wallet Page to see an insiders trading habits
Hidden features if not authorized by Auth0 login

Inspiration

Our team has been interested in the prediction markets and their predictive power for a long time. Seeing that Polymarket's data is entirely public, we decided to investigate various ML methods and data science techniques that can identify insider trading on the world's largest prediction market.

What it does

Our website streams Polymarket's trading data and curates data for inference. We apply a model, which we trained, to the incoming data to predict which of the many bets on the exchange are likely to have been placed by an insider.

We offer the insights on a web app to users who sign up (usinging Auth0 for authentication). These include rankings of recent insiders (ordered by probability of being an insider) and recent insider trades.

How we built it

We used the Polymarket API and custom scrapers to extract data, computed features and wrangled data in Polars, and built the model using XGBoost and scikit-learn. We used Fast-API and SQLAlchemy for our backend, and Vite + React for our front-end.

Auth0!

We fully integrated Auth0 into our app functionality. Services are hidden behind authentication, with blurred content when not signed in. Backend API endpoints are also protected by authentication.

Challenges we ran into

Working with data of this size was challenging, and we had to figure out work-arounds for computations that exceeded the RAM capacity of our machines. Optimizing our data operations was a great learning experience.

Accomplishments that we're proud of

In our team of three, two of us were participating in a hackathon for the first time (and our more experienced member was only on his second). We're proud of our time management and dedication during the event, and we can all say that we're very happy with the performance and design of the final product.

What's next for InsideRadar

More sophisticated features will be built in the future, to inform model development. The time (and cost) that it would take to scrape such features was a limitation of the 24 hours we had at this event.