-
-
Lens web application
Watch our video here: https://youtu.be/2jUSwG4XlPI
Inspiration
We were inspired by the success of Spotify and their wrapped product people love sharing data-driven narratives when they’re personal, fun, and insightful. But when it comes to personal finance people are more sensitive. We wanted to bring the same feeling to a banking app by making Bunq lens; a tool that makes your spending feel human and shareable — like a story only you could tell.
What it does
The Lens Agent connects to a user's Bunq account, analyzes a the last year’s worth of transactions using Llama-based AI models and NVIDIA acceleration, and generates a personalized financial recap. It highlights spending patterns, top categories, quirky purchases, and behavioral insights. Everything is presented via a web application as a narrative visualization that can accessible that can be shared socially.
How we built it
We used the Bunq API to securely fetch user transaction data; while we didn't quite get to the step of adding our own transactions, you can see our attempts on the GitHub. The backend, built with python organizes the data before sending it to a Llama-powered AI agent running on NVIDIA tokens. The agent interprets spending behaviors, identifies trends, and generates text summaries. On the frontend, we used React and Tailwind CSS to build an interactive, scrollable timeline.
Challenges we ran into
We faced several challenges in ensuring the AI-generated insights were both accurate and meaningful. Financial data is deeply personal, so striking the right tone was important. Working with a relatively new and unkown API like Bunq was quite tricky, which required careful handling of authentication and rate limits. Building a secure, privacy-respecting system that still felt personal and shareable took several iterations. Finally, crafting a user experience that felt like a story — not a dashboard — was a creative design challenge.
Accomplishments that we're proud of
We’re proud of turning raw transaction data into something that feels emotionally resonant. Users don’t just see numbers — they see themselves. We built a system that combines narrative, design, and AI in a lightweight, shareable package. And we did it in a hackathon weekend. The final product feels polished, and early testers were surprised by how seen they felt by their Wrapped.
What we learned
That API software maintenance is tricky for supporters but critical for developers! Also people are more likely to engage with and reflect on their finances when the experience is personal and even entertaining. We also deepened our understanding of prompt engineering for LLMs in high-sensitivity domains like finance, and gained valuable experience integrating third-party APIs securely in a fast-paced development cycle.
Limitations of the Implementation
The demo web application does visualize data which is retrieved from the agent's llama calls however, this is the most seamless part of the code. The brain of the agent is currently split into two poorly connected halves with non-matching data formats, the part which exports makes API calls to llama is not properly connected to the part that makes API calls to Bunq. Finally, we initially aimed to add our own data for testing purposes but this was not a priority and our stumbling blocks with the Bunq API including the delay with SugarDaddy were enough for us to focus on the other parts of the project as we got short on time.
Log in or sign up for Devpost to join the conversation.