We are a team from an insurance company in Japan. We are thinking of creating a smart contract that detects cyber crimes automatically to enable automated decentralized insurance in the future. The project was one component of this. We focused specifically on flash loan cyber crime. We tried to find a good way to automatically detect any kind of flash loan cyber crime. This project starts with retrieving on-chain data of AAVE, and analyzing the data to find signals of cyber crimes. We found out that, 1. Hackers use large amount of lending at cyber attacks, and 2. Hackers use brand new wallets without history so that they don't get detected. In the future, we'd like to implement solutions to automatically detect flash loan cyber crimes (and maybe even more) to pay out claims for our up-to-come decentralized insurance.
Retrieving on-chain AAVE data was done by "The Graph"'s GraphQL. We used hosted services that AAVE already had on The Graph, and used bubbletea repository to hit it from Python. The analytics were fully done by Python's Pandas framework. We originally were considering using the Dune Analytics dashboard to work on this analytics, but changed to GraphQL as it was more convenient to our use case.