Inspiration
As an engineer in fraud / risk at a fintech company, we employ a lot of custom machine learning models to analyze financial data and help protect our platform from bad actors and keep customer's money safe.
I thought it would be interesting to see how a solution like this might work in the crypto world. Much of the value of crypto / blockchain is derived from the decentralized and anonymous movement of digital currency but there have been a lot of high profile cases of scams and hacks on different blockchain platforms.
People that live in developing countries in Africa, Asia and South America are increasingly turning to crypto to help hedge against highly volatile and often rapidly devalued local fiat currencies. It also helps them send / receive money across borders faster and cheaper than traditional transfer services. A tool like this could help them verify that the blockchain addresses they're interacting with are legitimate.
What it does
Enter a blockchain address (currently only works for bitcoin) into the search bar at the top and click Analyze. The app will fetch all of the transaction data for that address and feed it to the OpenAI GPT model to detect any possible fraud patterns.
Try it out with some bitcoin addresses that have been involved in known exploits:
- bc1qnpc7u2ha7ct9c458rrqsawylz9e9j6jvkvzttt
- bc1q2skp47p9f5mr4n4m27k66v0l68gh3xdd7ad4e5
- bc1qtntflhe6f58lenv8chwuw8w55wxudwsvv549ah
How we built it
Built using the Next.js React framework and deployed to the cloud using Vercel.
Challenges we ran into
The main challenge was fine tuning the model prompt to get response output that makes sense and is more specific rather than being too general.
Accomplishments that we're proud of
Had a short timeline to complete this project due to finding out about the hackathon only a few days before the due date, so proud that I was able to deliver a working prototype.
What we learned
Previously attempted to build an AI assisted app where the models were self-hosted and that ended up being pretty cost prohibitive. This approach showed that interacting with the AI models through the OpenAI API is much simpler and the cost is actually very manageable.
What's next for crypto-guard.ai
- update it so that it also works for ETH addresses
- display more detailed blockchain address information
- fine tune the prompt to give better output
Log in or sign up for Devpost to join the conversation.