Inspiration
A hyped post in a Telegram group, Discord server, or TikTok remark offering 100x profits, urgency, and insider access is typically how young people and first-time cryptocurrency purchasers encounter frauds. The safety tools that are now available (RugCheck, Token Sniffer, GoPlus) are designed for seasoned traders. They respond with technical terms like "LP unlocked" or "mint authority active" and anticipate a contract address. The least equipped to utilise the technologies that could safeguard them are those who are most vulnerable to scams.
What it does
First-time cryptocurrency buyers have rug-pull protection. "A stranger sent me this coin" is transformed into a clear, fact-based danger report via Unruggable, which is designed to allow the AI to explain but never fabricate. After the user enters the coin's name, the trained AI system analyses its trends, including liquidity, the number of buyers, and whether or not the buyers are new accounts. Based on common patterns, the system generates a result that indicates a percentage of the coin's likelihood of being a scam.
How we built it
Our app was generated in Lovable, which we chose because it was one of the event sponsors. It’s built with React and TanStack and started from natural-language prompts, giving us a working web interface with an address checker, live data fetching, and a first version of the scoring model. Once we got our first model working, using solely our own processing, we combined it with the machine learning algorithm to get the current product.
Challenges we ran into
Throughout this project, we had numerous challenges. For instance, when we first started, we planned to use n8n for the automation of API usage, but this ran into numerous complications, so we had to make sacrifices by using direct accesses to them instead. Additionally, it was challenging to comprehend the problem breadth before beginning. Additionally, we encountered issues indexing token names when obtaining token data from third-party APIs; initially, we accessed them via token addresses, which made it challenging for the end user.
Accomplishments that we're proud of
We are proud of the uniqueness of our concept because we were keen to develop something that few other teams would have considered while still improving the lives of our users and benefiting the financial sector overall. Additionally, we executed our ideas fairly successfully and produced a project to a higher grade than we had anticipated.
What we learned
Due to n8n causing serious issues, we have learned to manage our time and resources effectively, making sure that we don't take on too much work for the limited time and that we use easily navigable resources. For example, finding data for Solana was difficult, and we found a lot of data for Erthium that was not currently relevant to us.
What is next?
Since the results are skewed and most of the coins come out as rug-pulls, we would train on a larger data set if we had more time. We would come across a user who had previously been conned and use their experiences—such as Telegram messages—as training material. Additionally, we would gather general input. The logic of the coin had inverted as a result of the training, being a byproduct of short training time. As a result, this is something we would look to fix in future publications of the software.