Garudpay: The "Undo Button" for Payments

Inspiration

₹18,000 lost in exactly 3 seconds.

That is the true story of our friend Vinayak. A scammer sent him just ₹1 to build trust, fabricated a fake story about a "mistaken transfer," and tricked him into sending back ₹18,000.

This type of manipulation is known as Authorized Push Payment (APP) fraud. It is the fastest-growing financial crime in the world today. The biggest problem with modern payments (both banks and crypto) is that they are instant and irreversible. Once you hit "Send," your money is gone.

We realized that educating users isn't enough because scammers manipulate people into a state of panic. We needed a systemic safety net. We needed an "Undo button" for payments. That's why we built Garudpay Escrow.

What it does

Garudpay Escrow is an AI-powered payment safety layer. It acts like an intelligent buffer between you and the person you are sending money to.

When you make a payment, our system does the following:

  1. AI Threat Detection: Our local AI acts as a watchdog. Before any money leaves your wallet, the AI instantly scans the transaction for common scam patterns (like extreme urgency, newly created wallets, or "accidental drop" tricks).
  2. The Safety Vault: If the transaction looks safe, it goes through instantly. But if the AI smells something phishy, the funds are not blocked entirely. Instead, they are sent to a secure, time-locked "Escrow Vault" on the blockchain.
  3. The "Undo" Window: While the money is locked in the vault (for a period of 4 to 24 hours), the sender has time to breathe and think. If they realize they are being scammed, they simply press the "Undo" button to instantly recall all their funds back to their wallet. No customer support needed.

How we built it

We built a robust, full-stack application divided into four main pieces:

  • The Mobile App (React Native/Expo): We wanted the user experience to be seamless, so we built a mobile-first wallet. This is where users send money, get scam warnings, and use the "Undo" button.
  • The AI Brain (Python & FastAPI): We used Python and LangChain to build an AI threat engine that analyzes transaction patterns in real-time to assign a risk score.
  • The Smart Vault (Solidity): We wrote secure Ethereum smart contracts to act as our Escrow Vault. This guarantees that the locked money is safe and the rules cannot be tampered with.
  • The Web Dashboard (Next.js): We built a web platform for managing the vaults and viewing system analytics using Next.js and Tailwind CSS.

Challenges we ran into

  • Balancing Security and User Experience: We didn't want to block transactions outright, as false positives ruin the payment experience. Figuring out how to use an "escrow" system instead of blocking funds was a major breakthrough for us.
  • Smart Contract Security: Writing the VetoVault smart contract was tricky. We had to ensure that the time-locks were perfectly calibrated and that unauthorized users couldn't drain the escrowed funds.
  • Connecting Mobile with Blockchain: Bridging a React Native mobile app with Ethereum smart contracts and a Python AI backend required careful architecture and seamless API integration.

Accomplishments that we're proud of

  • We successfully built a working "Undo" feature for blockchain payments, which natively goes against the traditional "irreversible" nature of crypto.
  • Our AI engine accurately flags manipulative "Penny Drop" scam patterns in real-time.
  • We created a gorgeous, intuitive mobile interface that hides all the complex blockchain and AI operations under a simple "Send" or "Undo" button.

What we learned

  • Education isn't enough to stop fraud; the underlying system design needs to protect users during moments of panic.
  • Smart contracts can be used for consumer protection, not just decentralized finance (DeFi) trading.
  • Combining AI behavioral analysis with the immutability of blockchain creates an incredibly powerful security paradigm.

What's next for Garudpay Escrow

We plan to expand our AI training models to cover a wider, constantly updating database of scam typologies. Furthermore, we want to build SDKs so that other existing traditional banking apps and crypto wallets can simply plug in our "Escrow layer" to protect their own users natively.

Built With

Share this project:

Updates