Moneymind

Most personal finance and budgeting apps share a fatal flaw: they are strictly retroactive. They send you a notification on Sunday to tell you that you blew your budget on Friday. That does nothing to change spending behavior in the moment. On top of that, managing multiple accounts and remembering reward categories is a headache: Does this cafe count as dining or grocery? Which card earns 3% versus 1.5% here? MoneyMind was born out of the frustration of standing at checkout guessing which card to swipe, and the realization that financial advice must happen proactively at the point of sale to truly prevent impulse spending and optimize rewards. What It Does Detects Real-Time Context: Uses GPS and the Google Places API to identify the merchant and commercial category (dining, groceries, retail) the user is currently visiting. Smart Card Optimization: Cross-references the location category against the user’s connected cards to instantly recommend the highest-earning card (e.g., 3% cash back on dining vs. 1% flat). AI Behavioral Coaching: Synthesizes account balances, recent transaction volume, and location context to deliver concise, real-time spending insights via an integrated AI model. Deterministic Safety Net: Implements a fail-safe architecture ensuring that even under API connectivity loss or rate limits, algorithmic fallbacks immediately step in without service disruption. How It Was Built Frontend: React, TypeScript, Vite, and Tailwind CSS for a responsive, modular financial dashboard. Backend: Python and FastAPI providing low-latency REST endpoints for location resolution and account aggregation. APIs & AI: Google Places API for venue discovery; Anthropic Claude 3 Haiku for contextual behavioral analysis. Demo Architecture: Injected a reproducible "Demo Mode" (Times Square coordinates) into the frontend to demonstrate live location-based category resolution reliably during judging. Challenges & Accomplishments Latency & Resiliency: Integrating live map queries and generative AI into a single synchronous request loop created latency and API failure risks. Solving this by engineering a custom backend "Safety Net" that catches AI timeouts and falls back to deterministic rule-based card mapping was a major milestone. Execution: Taking a full-stack idea from scratch to a working, crash-proof prototype with mapping and AI integrations within a 36-hour hackathon sprint.