Crypto that sits in a lending pool earns yield, but the moment you want to pay for something with it, you have to pull it out, swap it, and send it, by hand. DeQRypt makes it two moves: scan a QR invoice and pay straight from your DeFi position. Alongside it, an AI agent helps put idle crypto to work, without ever moving it on its own. We built it for the Base Hackathon.
- Hackathon
- Base Hackathon
- Team
- 5 people
- Network
- Base Sepolia
Scan to pay
Scan
The merchant's QR invoice.
Pay with IDRX
From the funds in your DeFi position.
Router contract
Swaps and settles in one atomic transaction.
Settled
The merchant is paid; nothing half-done.
The agent
Ask in plain language, like "invest my idle ETH", and the agent works out what you mean with LangChain and Gemini. It then looks up live APYs for trusted lending pools on Base, such as Moonwell and Aave V3, from DefiLlama, and recommends one for your token, with what you'd earn over two months, six months, and a year.
Your request
In plain language.
LangChain + Gemini
Reads the intent.
Live APYs
Trusted Base lending pools, from DefiLlama.
Recommendation
A pool and the projected profit. You decide.
Consultant mode
An agent that can move money is only as safe as what stops it. In consultant mode, the agent never lends on its own: it can only recommend. That rule lives in two places, the prompt and the code, so even if the model is talked into trying, the backend won't let the move through.
My part
I was one of the two AI agent architects and backend engineers on a team of five. I orchestrated the agent's flow, from reading the request to returning a recommendation, and built the API the app needed around it.
Where it ended
We didn't make the finals. DeQRypt runs on Base Sepolia, Base's test network, with test tokens; moving to mainnet and real protocol integrations was the next step on the roadmap.