bunq Guardian

A proactive AI layer on top of banking that listens, analyzes, and protects your budget in real time.

Comment

Inspiration

We were inspired by the idea that current banking apps are reactive — they show what already happened, but don’t help you before making a decision. We wanted to build an AI “guardian” that understands intent and actively protects users from overspending or risky transactions.

What it does

bunq Guardian is a multimodal AI assistant that:

  • Converts voice commands into financial actions
  • Scans receipts and categorizes spending automatically
  • Tracks user-defined budgets and warns before overspending
  • Classifies transactions using AI and shows how they impact your budget in real time

How we built it

We built a full-stack system using:

  • Frontend (HTML/CSS/JS) for a mobile-like UI and interaction
  • FastAPI backend for handling requests and routing logic
  • Whisper (Hugging Face) for speech-to-text
  • Claude API for reasoning and converting inputs into structured JSON
  • bunq Sandbox API for real banking integration (balance + payments)

The pipeline is: Voice/Image/Text → AI processing → Structured JSON → bunq API execution or analysis

Challenges we ran into

  • Integrating the bunq API (authentication, sessions, request structure)
  • Making AI outputs reliable and structured (JSON consistency)
  • Handling async frontend-backend communication smoothly
  • Debugging edge cases where AI classification didn’t match expectations
  • Keeping the UI responsive while multiple API calls happen

Accomplishments that we're proud of

  • Building a true end-to-end system (AI + frontend + real banking API)
  • Successfully combining multiple modalities (voice, image, text)
  • Creating a proactive budgeting feature, not just tracking spending
  • Getting real-time balance updates directly from bunq API
  • Turning natural language into actionable financial operations

What we learned

  • How to design systems around LLMs with structured outputs
  • The importance of validating AI outputs before execution
  • Practical API integration and debugging in real-world systems
  • Managing state and async logic in frontend applications
  • How multimodal AI can be applied to real-life use cases

What's next for bunq Guardian

  • Smarter budgeting with adaptive AI recommendations
  • Personalized spending insights over time
  • Stronger fraud detection using behavioral patterns
  • Full production integration with bunq (beyond sandbox)
  • Mobile app version for real-world deployment

Built With

Share this project:

Updates