GivingGenie
Snap a receipt, earn points, gift them to a friend. No app, no wallet, no seed phrase.
Project Description
An AI agent (Telegram today, SMS-ready) that runs an on-chain cashback program for small businesses. Customers earn USDC-backed loyalty points by photographing a receipt or scanning a QR code, then check, redeem, or gift those points — all in a chat thread, with a custodial smart wallet provisioned silently behind the scenes. Every gift carries the business's name, turning each redemption into organic acquisition.
How it's Made
LoyalChain is built as an AI-first loyalty platform with a modular agent architecture. Customers interact through a Telegram bot (designed behind a transport adapter so SMS can be added later without changing the core logic). Messages are normalized and sent to an agent service where Claude Sonnet performs natural-language intent parsing and receipt OCR, while the backend independently validates every action before execution to prevent AI hallucinations from moving funds.
On-chain logic is implemented in Solidity using Hardhat and OpenZeppelin, with two smart contracts deployed on Arc Testnet: an ERC-1155 LoyaltyPoints contract for merchant-specific reward points and a MerchantEscrow contract that atomically burns points and releases USDC from a merchant-funded escrow pool. Dynamic Server Wallets (MPC) provision custodial wallets automatically, eliminating seed phrases and wallet onboarding friction for end users.
Supabase stores customer records, pending actions, receipt claims, and wallet metadata, while viem provides blockchain interactions and live ENS resolution so users can gift points directly to .eth names. A dedicated Ethereum mainnet client resolves ENS names in real time before transactions execute on Arc.
One of the more hacky—but effective—parts of the project was separating AI decision-making from financial execution. Claude only proposes structured intents; the backend re-validates balances, recipients, affordability, and idempotency before requiring an explicit confirmation step and signing transactions. This let us deliver a conversational UX while maintaining deterministic, secure transaction execution.
