Inspiration
We were inspired by the gap between financial data and actual understanding. Traditional banking apps—like bunq—show transactions but don’t answer the real questions users have: “Am I overspending?” or “Can I afford this?”
We also noticed how paper receipts are still disconnected from digital finance, and how navigating banking apps feels like work instead of conversation. That led us to rethink finance as something you talk to, not something you click through.
What it does
bunqAFA (bunq AI Financial Assistant) is a single-page AI-powered finance interface that combines:
Receipt scanning (turns paper receipts into structured transactions) Spending overview (categorized insights, charts, and summaries) Smart savings (AI-recommended deposits) Natural-language chat (ask questions like “Am I okay this month?”)
All of this is wrapped in an interactive 3D interface built with Three.js, where users can ask instead of navigate.
How we built it
We built bunqAFA as a full-stack app with a local AI layer:
Frontend: Vanilla JS + Three.js for a real-time 3D UI Backend: Node.js + Express + TypeScript AI layer: Ollama running a local LLM (llama3.2) Banking: bunq sandbox API with real account simulation
We used a hybrid AI system:
Fast keyword matching for common commands LLM-based intent classification for flexible language Structured JSON prompts for consistent outputs
Everything runs locally with a one-command setup script, making the project fully reproducible for judges.
Challenges we ran into OCR reliability: Receipts vary wildly in format, quality, and language Latency vs UX: Keeping AI responses fast enough to feel real-time Intent detection: Mapping messy natural language to precise actions bunq API complexity: Handling authentication, sessions, and RSA signing State synchronization: Keeping scanned receipts, mock data, and real transactions consistent 3D UX balance: Making the Three.js scene meaningful, not just decorative Accomplishments that we're proud of A fully working end-to-end product (not just a prototype UI) Seamless integration of image (OCR), audio (voice), and text (AI chat) A local-first AI system using Ollama (no external APIs required) A clean, opinionated UX that reduces finance to 3 core actions A one-command setup that installs everything and runs instantly Turning a 3D interface into a functional UX element, not just visuals What we learned AI is most useful when it’s embedded in workflows, not added on top Simplicity beats flexibility in financial UX Local LLMs (via Ollama) are powerful enough for real applications Combining modalities (image + voice + text) creates a much more natural interaction model Good defaults and fallbacks are essential—AI should never block the UI What's next for bunqAFA Smarter categorization using learned behavior instead of static keywords Real-time notifications and proactive insights Multi-month trend analysis and forecasting Deeper integration with bunq accounts (beyond sandbox) Mobile-first experience with improved voice interaction Personalization: adapting advice based on user habits
Ultimately, we want bunqAFA to evolve from a financial dashboard into a true financial companion that understands, predicts, and guides user behavior in real time.
Log in or sign up for Devpost to join the conversation.