Both of us are quite new to the blockchain space. Since we have backgrounds in ML we knew we wanted to create some sort of application that bridges the two areas. After talking to people in the space we discovered that there might be an opportunity to help AAVE using ML to determine the credit of potential borrowers. After looking more into this we decided that we wanted to build it out, and now here we are!

What it does

CredAAVE evaluates whether or not an account is likely to default on its AAVE loan based on their previous account activity on AAVE. CredAAVE essentially acts as a credit score for AAVE.

How we built it

We mainly used python to build out the actual code and machine learning used to determine the end result. We also used GraphQL to get the training data of AAVE users (including those who've defaulted on their loans in the past), and at inference whenever you enter the ETH address of the borrower.

Challenges we ran into

There were loads of challenges we ran into from the very beginning with figuring out how to get the data from Graph to the end with connecting the front end with the back end.

Accomplishments that we're proud of

We're the most proud of coming up with a practical and original idea to leverage ML for blockchain and actually creating it! We did not have that much experience in blockchain so getting over technical barriers was quite difficult - but very rewarding in the end.

What we learned

We truly learnt a ton about bridging the gap between blockchain and ML - something that has not been done very often. The abilities to figure things out, time management, and team organization were also crucial for this hackathon - and were abilities we definitely developed during it. On the technical side, we also learnt a ton about using Graph, making a front end, connecting the back end to a front end, and of course creating an ML model in the blockchain space.

What's next for CredAAVE

We want to really flesh out this project! This would involve getting even more data to train the model on, improving the front end, and launching a web app for it!

Built With

Share this project: