-
Checking the balance in our app.
Inspiration
Globally, there are 43 million blind people and over 2 billion with vision impairments. With so much of financial management being digital today, we realized people with these visual impairments likely struggle to manage their accounts and payments. In addition, many elderly people may have issues navigating these digital apps and may be confused by all the features. That’s why we decided to build Strombus, a voice agent that simplifies everything for the user and is controllable fully by audio. The name Strombus is inspired by the voice of the ocean you hear when you put your ear to a conch, which belongs to the genus Strombus.
What it does
Strombus accesses your bank information, but lives behind the normal protection you would have in your banking app such as facial recognition. This way, not anyone can access your financial information by simply calling it into existence. The user just has to press the microphone on the screen (which is near the center for easy access) and begin talking. It can ask Strombus to check their bank balance, to search for any unpaid bills, and to pay those bills; all without needing to navigate an app.
How we built it
We utilized ElevenLabs agents, Capital One’s Nessie, a custom MCP that interacts with Nessie, and Vercel for deployment to bring this to life. We also used AI such as Claude Code to help us with its development and coding.
Challenges we ran into
Getting the voice agent to behave well took a lot of iteration. Its responses sometimes sounded stiff or ran too long to be useful when spoken aloud, so we reworked how it phrases things for listening instead of reading. It also struggled to understand some user input, especially when people asked for the same thing in different ways or paused mid-sentence. The hardest part was moving money: getting transfers to work with Nessie through our custom MCP took a lot of trial and error. Each bug showed us where things break, and we debugged our way through them so those mistakes are less likely to come back.
Accomplishments that we're proud of
We got the voice agent working end to end. A user can check their balance, find unpaid bills, and pay them, all by voice. We're especially proud that the backend and frontend work together smoothly, so it feels simple even for someone who doesn't usually use banking apps.
What we learned
Debugging an AI agent is different from debugging regular code. A wrong answer doesn't always throw an error, so we had to trace the conversation, the tool calls, and the API responses to find where things went off track. It was tough, but we learned that building something like this is very doable if you break the problem down and analyze each piece.
What's next for Strombus - Your Banking Voice Agent
We want to make the UI even better, with clearer feedback so users always know what Strombus just did. We also plan to expand to more platforms and add more everyday banking tasks, so more people can manage their money without having to navigate an app.
Built With
- elevenlabs
- python
- render
- typescript
- vercel
