Wallet Whisperer

Inspiration

“Can we afford it?” sounds like a simple question—until you look at your account.

Your checking balance might look healthy, but rent is due before your next paycheck. Your savings might cover a purchase, but some of that money already belongs to an emergency fund or another goal. A balance alone does not mean you can buy something.

Planning with other people makes the question even harder. Friends organizing a trip or families making a shared purchase need to discuss contributions without having a public review of everyone’s finances.

We created Wallet Whisperer to connect those two problems: understanding what your money can buy and making shared spending decisions with less uncertainty and social pressure.

What it does

Wallet Whisperer brings accounts, bills, debts, income, and savings goals into one financial picture, then helps users explore their next decision.

Our prototype includes three core experiences:

My Money: A consolidated dashboard with cash-flow forecasts and five explainable financial health indicators: essential bills covered, safety cushion, debt load, savings protection, and nonessential subscriptions.
“Can I buy it?”: A purchase simulator that considers the cost, date, and exact account being used. Users can see how a purchase affects their financial health before spending, with outcomes such as “Yes,” “Yes, if…,” or “Not right now.”
Huddles: A shared-purchase workflow where participants pledge contributions, receive individual affordability checks, review revealed amounts, and agree on a plan. The group view separates contribution information from personal account details and assessment feedback.

The goal is to make financial decisions understandable: meaning what changes, why it matters, and what options are available.

How we built it

We built the application with Next.js 16, React 19, TypeScript, and Tailwind CSS 4.

Capital One’s Nessie API supplies sandbox banking records, including accounts, transactions, bills, and loans. We normalize those records into a shared financial model that powers the dashboard and simulations.

The foundation is a deterministic calculation engine. It forecasts balances, evaluates financial health, checks funding sources, and calculates affordability. All money is represented as integer cents, and every proposed purchase names a specific source account.

We use MongoDB Atlas with the official MongoDB driver to persist application data, including health history, purchases, user accounts, social connections, and Huddles. Server-side sessions identify the signed-in user, while Huddle responses are filtered for that participant.

We also built Vitest regression tests around the financial engine. Our guiding rule is that financial numbers must come from reproducible calculations. The planned NVIDIA Nemotron layer will use those calculations to explain results and explore possible adjustments.

Challenges we ran into

One challenge was defining what “affordable” actually means. Having enough money today does not guarantee that upcoming bills remain covered. We had to account for timing, recurring obligations, savings allocations, and the account a purchase draws from.

Another challenge was avoiding misleading calculations. A recurring expense can appear in multiple data sources, and money assigned to a savings goal must not be counted as extra available cash.

Group planning introduced a different set of problems: deciding when pledges become visible, which changes require renewed agreement, and how to handle participants acting from different devices. That pushed us to model Huddles as explicit phases and use versioned database updates to handle competing changes.

Accomplishments that we're proud of

We’re proud of turning a broad budgeting idea into a focused decision tool.

The purchase simulator connects an everyday question to an explainable forecast. The health assessment makes its individual indicators visible instead of asking users to trust an unexplained score. Huddles bring the social side of spending into the same experience.

We’re also proud of building a reusable, testable financial engine that can support both the interface and future AI features.

What we learned

We learned that financial trust depends on details: dates, account ownership, missing information, and the difference between available money and money already set aside.

We also learned that privacy affects the entire product. It shapes what the server returns, when information is revealed, and how a group discusses changes.

Most of all, we learned that useful financial guidance needs a clear explanation grounded in consistent calculations.

Additionally thinking about User flow when creating our UI and design.

What's next for Wallet Whisperer

Our next priority is integrating NVIDIA Nemotron to explore engine-verified scenarios and explain practical options, such as delaying a purchase or adjusting discretionary spending.

We also plan to strengthen explicit household-sharing permissions, improve onboarding beyond the seeded banking profiles, and complete provider-confirmed sandbox funding for approved Huddles.

Longer term, we want private alerts when changing finances put a planned purchase at risk, along with richer comparisons between possible plans.

Our vision is simple: help people understand their money, discuss shared goals comfortably, and answer “Can we afford it?” with a plan they can explain.