Splitsies

• 

  App Load In

What it does: Splitsies makes splitting bills effortless. Splitsies lets you snap a photo of a receipt and automatically itemizes it into a clean, editable list so you can split rent, food, and groceries without manual math. You can assign items to people or use percentage and exact-amount splits, then send each person a payment request via phone number.

How we built it:

Built an iOS app in Swift with a smooth, approachable UI focused on fast scanning and quick edits.

Uses an AI parsing flow to convert receipt images into structured, itemized data with a strict schema for easy review and correction.

Integrates contacts and phone numbers to generate per-person requests that users can send via their preferred payment method, including Zelle via phone where supported.

Employed prompt engineering and schema validation to improve extraction reliability and reduce malformed outputs.

Core features:

Picture Receipt: Take or import a photo and get an itemized bill you can edit quickly.

Sort Tabs: Split by item, percentage, or exact amounts; assign who covers whom; handle refills and cash payers.

Zelle/Number Send: Select friends by phone number and send each their total with a prefilled request message.

Payment History/Dashboard: Track requests sent and settled with itemized context.

Challenges we ran into:

Getting consistent AI itemization from varied receipts, where some calls returned incomplete or merged lines.

Prompt sensitivity required iteration and validation to stabilize outputs.

Accomplishments that we're proud of:

A smooth, fun UI that makes splitting feel fast instead of stressful.

An itemized, editable receipt structure that keeps users in control with minimal taps.

Clear, contact-based payment requests that work regardless of the recipient’s preferred payment app.

What we learned:

Swift app architecture and lifecycle patterns for camera, permissions, and async data flows in a real project.

Prompt engineering, schema design, and guardrails dramatically improve AI extraction reliability.

Future Plans:

Voice-based “how to split” commands as a future enhancement once reliability meets the bar.

Multi-model fallback and confidence scoring for better accuracy on messy receipts.

Built With

• figma
• gemini
• swift