Inspiration

Most people keep their money in many places: checking, savings, a CD, a retirement account, a credit card and a car loan. Few of them can see how they're actually doing, or what to do next. I came into this weekend as a first-time hacker working solo, and that feeling of having too many things to track and no clear next step is exactly what I wanted Compass to fix.

What it does

Compass is an AI financial co-pilot. It shows one customer's whole financial picture on one screen and coaches them on their next move.

  • Accounts overview: checking, savings, CD, Roth IRA, credit card and auto loan, with net worth
  • Financial health score (0–100): built from four indicators: emergency fund, credit utilization, debt-to-income and savings rate, each explained in plain English
  • 12-month forecast: projected net worth, cash and debt
  • "What if" simulator: sliders for cutting dining out, paying extra on the credit card, or changing income, with the forecast and score updating live
  • AI coach (Gemini): a chat that reads the customer's real numbers and the current what-if settings, so answers match what's on screen
  • Toyota "next move" card: compares the 6.4% car loan to the 24.99% credit card and recommends the single best next step (pay the card first), with the estimated interest saved

How I built it

React 18, Vite and Recharts on the front end, with an Express server that keeps all API keys off the browser. The server calls the Google Gemini API for the coach and Capital One's Nessie API for customer data (checking, savings and credit card from Nessie accounts, the auto loan from Nessie loans, and rent, utilities and subscriptions from Nessie bills). Responses are cached, and the app falls back to demo data and offline coach answers if an API is unavailable, so it never breaks. All the money math lives in plain functions shared by the score, forecast, simulator and coach.

Challenges I ran into

  • Making sure the AI coach, the charts and the score all use the same numbers
  • Mapping Nessie's account types (Checking, Savings, Credit Card) to a fuller financial picture
  • Handling API failures gracefully

Accomplishments that I'm proud of

Built a complete, working product solo, as my first hackathon.

What I learned

How APIs, AI and money math fit together, how to keep API keys out of a public repo, and that a small finished project beats a big unfinished one.

What's next

Pull spending from Nessie transactions, add goals and alerts, and let customers connect more account types.

Tracks

PNC (360° Financial Health Navigator), Capital One (Best Use of Nessie), Toyota Financial Services (What's Your Next Move?), MLH Best Use of Gemini API

Built with AI assistance (Claude Code). All data is synthetic.

Built With

Share this project:

Updates

Submission history