Bunq Spending Advisor

Bunq Spending Advisor is an advisor where AI assistant Finn automatically categorizes transactions, detects subscriptions, and forecasts your month-end balance to build smarter spending habits.

Comment

Inspiration

We noticed that most people have no idea where their money actually goes. Bank apps show raw transaction lists, but they don't tell you why you're broke by the 20th. We wanted to build something that feels less like a spreadsheet and more like a smart friend who happens to know your finances proactive, honest, and a little witty.

What it does

bunq Spending Advisor connects to your bunq account via real-time webhooks and gives you a full picture of your financial life:

  • Detects subscriptions you might have forgotten about, and shows the annual cost
  • Forecasts your month-end balance based on your spending pace and budget
  • Finn AI chat - ask anything about your finances in plain language and get short, actionable advice
  • Receipt scanning - upload a photo of a receipt and get itemized breakdown and saving tips powered by Claude's vision

How we built it

  • Frontend: React + Vite + Tailwind CSS — dark-mode mobile-first UI with tab navigation (Home, Subs, Forecast, Finn AI, Receipt)
  • Backend: FastAPI (Python) with SQLite for transaction storage
  • AI: Anthropic Claude API for subscription detection, financial chat (Finn), and receipt image analysis
  • Banking: bunq Sandbox API + real-time webhooks to receive payment events instantly
  • Infrastructure: ngrok for local webhook tunneling during development

Challenges we ran into

  • Webhook reliability: Making sure every bunq payment event was captured, deduplicated, and stored correctly in real time
  • Merchant name extraction: bunq transaction counterparty fields are often messy; we had to use Claude to clean and normalize merchant names
  • Forecast accuracy: Building a month-end balance forecast that accounts for recurring subscriptions, spending trends, and user-set budgets without overfitting to short history
  • Receipt OCR edge cases: Receipts come in wildly different formats; getting Claude's vision to reliably itemize them took careful prompt engineering

Accomplishments that we're proud of

  • End-to-end real-time pipeline: a bunq payment triggers a webhook, which hits our backend, runs Claude categorization, and appears in the UI within seconds
  • Finn AI feels genuinely useful - short, emoji-punctuated answers that don't waste your time
  • Receipt scanning that actually works on crumpled, angled, or low-light photos
  • Subscription detection that surfaces annual costs, making the "death by a thousand cuts" problem visible at a glance

What we learned

  • How to integrate bunq's SDK and webhook system end-to-end
  • How to use Claude's multimodal API for real-world image understanding
  • Prompt engineering for financial advice: brevity and specificity matter far more than length
  • How to build a useful forecast with limited transaction history

What's next for bunq Spending Advisor

  • Savings goals: Let users set a goal (e.g. "save €500 for a trip") and have Finn track progress and nudge spending behavior
  • Peer benchmarking: Anonymous comparison of spending patterns with similar users ("You spend 2x the average on food delivery")
  • Smart alerts: Proactive push notifications when spending in a category is trending over budget
  • Multi-account support: Aggregate across multiple bunq accounts or sub-accounts
  • Production bunq API: Move from sandbox to live accounts for real users

Built With

Share this project:

Updates