Venue Rewards Engine
A receipt-to-reward platform for hospitality venue staff — a waiter photographs a receipt, AI reads it and matches qualifying sales, and the system rewards them automatically while an owner only ever sees the exceptions.
Manual receipt-checking doesn't scale, and it's easy to game.
A venue wants to reward staff for driving sales of a specific product line, but checking every receipt by hand doesn't scale past a handful of venues — and once there's a cash reward attached, duplicate and fabricated receipts become the first thing anyone tries.
AI extraction, fraud layers enforced at the database.
A waiter photographs a receipt; an AI extraction step reads the line items and matches them against the venue's reward rules; low-confidence matches route to manual review instead of auto-approving. Every fraud defence sits at the database layer, not just in app logic.
AI reads the receipt image and matches sales against per-venue reward rules, with confidence-based routing to a review queue.
Insert-time duplicate and fingerprint checks, a decaying risk score, auto-suspend and bank-detail reuse detection, all enforced in the database.
Bank details stored encrypted at rest, written only through a locked-down RPC, never read back in plain text.
A desktop dashboard plus a mobile "today" view with swipe-to-approve, so the owner only handles what the system flagged.
The hard part is built and self-checked; launch is next.
The security model is the part that had to be right before anything else: a row-level-security self-check covering waiter isolation, self-approval and self-unsuspend blocks, and anonymous lockout has been run live against the real schema and passed with zero leftover rows. The receipt pipeline, fraud layers, rewards ledger and both owner dashboards are built; what's left is store listing, legal review and production hosting — not re-engineering the core.