WHOOSH — Split Now. Not Later.

Inspiration

We've all experienced it: a group of friends wants to order dinner, book a trip, or buy concert tickets, but someone has to put the entire purchase on their card.

Everyone promises to pay them back. Some people transfer the money immediately. Others forget, need reminders, or take days to reimburse the organizer.

The idea for WHOOSH came from a real experience involving an Uber Eats order that resulted in approximately $200 being charged to one person's card. It raised a simple question:

Why does buying something together still require one person to pay for everyone?

Traditional expense-splitting tools primarily help people settle debts after a purchase. We wanted to explore a different approach: what if everyone could pay their share during checkout instead?

That's how WHOOSH began.

What It Does

WHOOSH is an AI-powered group checkout prototype that helps people coordinate individual payments toward a shared purchase.

Instead of one person covering the full amount and chasing reimbursements, WHOOSH allows participants to review their allocated contributions and complete separate PayPal checkouts.

The experience includes:

  • AI-powered splitting: Describe how a purchase should be divided in natural language, including exceptions and item-specific arrangements.
  • Individual PayPal payments: Each participant can approve and pay their own contribution through PayPal.
  • Payment verification: WHOOSH independently verifies successful PayPal captures rather than relying on participants to report that they've paid.
  • Funding progress: Organizers can see which contributions have been verified and how much remains outstanding.
  • AI Checkout Coordinator: A read-only AI assistant can inspect payment status and recommend next steps.
  • Double-entry ledger: Payment events are recorded through an accounting-oriented ledger model.

The goal is to make shared purchases feel like a coordinated checkout rather than a collection of personal debts.

How We Built It

WHOOSH uses a React and TypeScript frontend, a Node.js backend, and PostgreSQL for persistent data.

AI-Powered Split Interpretation

We integrated Muse Ember 1.0 to interpret natural-language instructions about how a group wants to divide a purchase.

For example, consider four friends booking a $1,284 weekend trip to Montreal, including traveler passes, a cabin upgrade, and a station shuttle.

The organizer can enter:

Sam isn't taking the shuttle. Jordan and Avery are covering the cabin upgrade. Split everything else equally.

The AI interprets these instructions and converts them into structured allocation rules.

Importantly, AI does not have the final say over financial arithmetic. Deterministic validation checks the resulting allocations at the cent level to ensure they are consistent with the purchase total.

PayPal Integration

We integrated the PayPal Orders v2 API in the sandbox environment to support separate payment flows for individual participants.

Each participant can complete their own PayPal checkout. WHOOSH then verifies the corresponding payment capture and updates the shared purchase's funding status.

This is a coordinated collection of independent PayPal payments, not a single atomic multi-payer merchant transaction.

AI Checkout Coordinator

We built an AI Checkout Coordinator that can use read-only tools to inspect the state of a group purchase, including verified payments and outstanding contributions.

The coordinator can explain the current funding status and suggest next steps without independently moving money.

Payment Records and Voice

We implemented a double-entry ledger model to support auditable payment records and integrated ElevenLabs voice transcription as an additional input capability.

Together, these components connect natural-language instructions, payment processing, and funding visibility into one workflow.

Challenges We Faced

1. Coordinating Multiple Independent Payments

One of the biggest challenges was designing a shared checkout experience around payment infrastructure that processes individual transactions.

We needed to distinguish between a participant approving a payment, PayPal confirming its capture, and the overall group purchase reaching its funding target.

We addressed this by tracking each participant's contribution separately and using verified payment information to determine funding progress.

2. Making AI Reliable for Financial Calculations

Natural-language instructions can be ambiguous, especially when purchases contain multiple items and different participants have different obligations.

A language model might understand what someone intends without consistently producing mathematically valid allocations.

To reduce this risk, we separated AI interpretation from deterministic financial validation, using cent-level calculations rather than trusting generated amounts without verification.

3. Maintaining Consistent Payment Records

A shared purchase involves several related states: allocations, payment attempts, successful captures, outstanding balances, and ledger entries.

Keeping these records consistent required careful backend design and a clear distinction between confirmed and unconfirmed contributions.

4. Balancing Ambition With Hackathon Constraints

WHOOSH combines payments, AI, backend orchestration, and user experience design. Bringing these systems together within a limited development window meant prioritizing a demonstrable end-to-end prototype over building every possible feature.

We focused on showing the core concept while being transparent about the limitations of the sandbox implementation.

Accomplishments We're Proud Of

  • Built a working group checkout prototype integrating PayPal sandbox payments.
  • Implemented natural-language, item-aware split interpretation with deterministic validation.
  • Developed independent payment verification and shared funding tracking.
  • Created an AI coordinator that can inspect purchase state and recommend next steps.
  • Implemented a double-entry ledger model for payment records.
  • Connected the technical implementation to a real-world problem through a product demo.

Most importantly, we turned a familiar frustration into a tangible product concept.

What We Learned

Building WHOOSH reinforced that payments are not just a user-interface problem.

A seemingly simple experience letting four friends pay for one purchase requires careful handling of transaction states, allocation accuracy, verification, and accounting.

We also learned that AI is most useful in financial workflows when its responsibilities are clearly defined. Language models can make complicated instructions easier to express, while deterministic systems handle the arithmetic and enforce financial constraints.

Finally, we learned how much work goes into communicating a technical product clearly, from designing the checkout experience to producing a demo that makes the problem understandable.

What's Next for WHOOSH

Our longer-term vision is to make group checkout a payment option that merchants and booking platforms could integrate directly.

Potential next steps include:

  • Exploring merchant integrations and production payment infrastructure for coordinated group purchases.
  • Supporting more complex payment arrangements, including backup contributors and failed-payment recovery.
  • Expanding natural-language splitting to recurring groups and additional purchase categories.
  • Conducting usability testing with real groups to validate the checkout experience.
  • Improving the AI coordinator's ability to explain funding states and guide organizers through payment exceptions.

WHOOSH is currently a hackathon prototype. Payments are processed through PayPal's sandbox as independent transactions, and merchant booking or fulfillment is simulated. Production merchant settlement would require additional infrastructure and integration.

Our vision remains simple:

No fronting the full bill. No chasing reimbursements. No becoming your friends' unpaid accountant.

WHOOSH — Split now. Not later.

Built With

Share this project:

Updates

Submission history