Inspiration

Every group dinner, hostel flat, or weekend trip ends with the same headache: someone holds a messy paper receipt, everyone argues over who ordered what, and the person who paid front-up spends days chasing friends down for money. We wanted to build an effortless tool where you just paste raw, unstructured conversational notes or chat messages, and AI + PayPal handles the math and settlement in seconds.

What it does

SplitAI transforms unstructured expense notes into clear financial distributions and frictionless payments:

  • Conversational Parsing: Users paste natural language notes (e.g., "Aman had a burger for $18, Sneha had pasta for $15, split $8 tip"), and Google Gemini parses who owes what.
  • Fair Tip & Tax Distribution: Automatically calculates shared expenses, discounts, tips, and taxes proportionally or equally.
  • Multi-Currency Support: Handles both USD ($) and Indian Rupees (₹) with custom financial logic (like flatmate rent and utility splits).
  • Instant PayPal Settlement: Generates one-click direct PayPal settlement links so users can pay their exact dues with zero friction.

How we built it

  • Frontend & UI: Developed a responsive, interactive web application using Python and Streamlit.
  • AI Engine: Integrated Google Gemini API with custom prompt engineering to extract itemized costs, handle calculations, and respond to custom user constraints.
  • Resilient Multi-Model Architecture: Built an automated model rotator across Gemini models (gemini-3.5-flash-lite, gemini-3.6-flash, gemini-3.8-flash) with an offline calculation engine to prevent rate-limit interruptions.
  • Payment Infrastructure: Connected PayPal Developer Sandbox tools and PayPal payment links for instant digital settlement.
  • Cloud Deployment: Hosted live on Streamlit Community Cloud and version-controlled via GitHub.

Challenges we ran into

  • Model Versioning & Quotas: Overcoming dynamic API rate limits across free-tier models by implementing auto-fallback and clean error handling.
  • Mathematical Consistency: Fine-tuning prompts to ensure the LLM calculated exact arithmetic without hallucinating extra charges or missing shared taxes.
  • Input Sanitization: Automatically stripping whitespace and formatting special characters from user payment handles so generated PayPal URLs never break.

Accomplishments that we're proud of

  • Built, tested, and deployed a fully functional FinTech application end-to-end as a solo developer.
  • Created a live, publicly accessible web app on Streamlit Cloud that anyone can test on mobile or desktop.
  • Successfully bridged Generative AI reasoning with practical PayPal payment workflows.

What we learned

  • How to interface with modern Generative AI SDKs and engineer production-ready prompts.
  • How PayPal Sandbox accounts and developer credentials simulate merchant-buyer payment flows.
  • Rapid full-stack deployment using Python, GitHub, and cloud platforms.

What's next for SplitAI - Smart PayPal Splitter

  • OCR Receipt Scanning: Allowing users to upload photos of paper receipts using computer vision for instant text extraction.
  • Automated PayPal Payouts: Upgrading from payment links to the PayPal REST Orders & Payouts API for direct multi-party automated wallet transfers.
  • WhatsApp Bot Integration: Enabling group chats to forward bills directly to an AI agent for instant group settlement.

Built With

Share this project:

Updates

posted an update —

As the solo developer, I built SplitAI completely from scratch end-to-end:

  • Frontend & UI: Designed and developed the interactive web application using Python, Streamlit, and custom PayPal FinTech CSS styling.
  • Generative AI Integration: Integrated the Google Gemini API with custom prompt engineering for conversational expense parsing, fair tax/tip distribution, and multi-currency math (USD and INR).
  • Payment Infrastructure: Integrated the PayPal Developer Sandbox environment and engineered dynamic, one-click settlement link generation.
  • DevOps & Deployment: Set up the GitHub repository, managed project documentation, and deployed the live production application to Streamlit Community Cloud.

Submission history