BUSINESS · HUSKYHACK · SOUND CREDIT UNION CHALLENGE
BANKS TRY SMALL MODELS FOR PERSONAL OFFERS
Smart savings that find the member — not the other way around
SEATTLE — Credit union members leave money on the table every day. Cashback offers go unnoticed, local deals expire unclaimed, loyalty points pile up unredeemed. Banks hold the transaction data to fix this; most simply do not act on it. Nudge, built for the Sound Credit Union challenge at HuskyHack, turns that history into a proactive savings engine — inside the app members already use.
The feature lives as a dedicated tab. It reads spending history to curate a marketplace of relevant offers — coffee shops, grocery cashback, travel rewards, loan promotions — filterable by category, sorted by relevance, and redeemable in-store by QR code.
HOW IT WORKS
Proximity does the persuading: when a member walks near a partner merchant, Nudge fires a real-time notification — "You're 0.4 miles from Stumptown Coffee — get 25% off now." No separate app, no opt-in friction; the right offer at exactly the right moment.
A savings dashboard charts five months of history with Recharts, breaks savings down by category, and projects what current habits are worth over time. GPT-4o-mini reads the member's actual savings profile and writes personalized tips — cheap enough to call per session, fast enough to feel live.
LESSONS FROM THE FLOOR
The small-model bet paid off: a compact LLM with the right context beat a large one without it, at a fraction of the cost. Location context changed everything about perceived relevance, and strict TypeScript caught data-shape bugs that would have burned scarce hackathon hours. Rendering the app inside a browser iPhone frame made demos instantly legible to judges.