Inspiration

As students, we lose track of small purchases: coffee, food delivery, subscriptions we forgot about. Budgeting apps ask you to tag every transaction yourself, so most people quit after a week. We wanted an assistant that does the boring part for you and just tells you what to change.

What it does

ReceiptMind lets you snap a photo of a receipt or upload a bank statement. The AI:

  • Reads and pulls out every item, price, merchant and date
  • Sorts the spending into categories automatically (food, transport, subscriptions, etc.)
  • Spots wasted spending, like repeat subscriptions, impulse buys and price spikes
  • Builds a simple weekly budget and gives plain-English tips, e.g. "You spent $42 on delivery this week; cooking twice would save about $25."

How we built it

  • Frontend: React web app with photo upload
  • Backend: Python (FastAPI)
  • AI: OCR and a vision-language model to read receipts; an LLM to categorize spending and write budget tips; a model fine-tuned with Tinker (Thinking Machines) to sort categories more accurately
  • Data: PDF/CSV parsing for bank statements, stored in a lightweight database
  • Hosting: Google Cloud

Challenges we ran into

  • Receipts are messy: crumpled paper, faded ink and different layouts for every store
  • Every bank formats its statements differently
  • Keeping the AI's tips specific and useful instead of generic
  • Treating financial data privately and securely

Accomplishments that we're proud of

  • An end-to-end flow from a receipt photo to a personalized budget
  • Spending categories that come out right without manual tagging
  • Advice that feels like a friend helping you, not a spreadsheet

What we learned

  • How to combine computer vision and LLMs in one pipeline
  • Prompt design and fine-tuning for structured financial data
  • Clean, simple UX matters as much as the AI does

What's next for ReceiptMind

  • Securely link bank accounts directly
  • Shared budgets for roommates
  • Savings goals and smart alerts
  • A mobile app with instant receipt scanning

Built With

Share this project:

Updates

Submission history