Conta Aberta
A car subscription app that shows customers what their monthly fee covers. It started at the Ruptura 2026 hackathon and I kept building it afterwards.
- Context
- Ruptura 2026 hackathon
- Started
- Type
- Full stack
By the numbers
- hackathon days to the first prototype
- 4
- .NET API endpoints
- 12
- automated tests for the calculation engine
- 14
- assumptions backed by public sources
- 7
- AI providers: Claude and OpenAI
- 2
Counted in the code in Oct 2026.
Architecture
The challenge
I spent four days at Ruptura 2026, inside Localiza Labs, on a challenge about the subscription app: get customers to open it in their daily routine, not only when something is wrong. In the first mentoring session I was asked what the pain was, and I couldn't answer. We started over and went looking for data.
The case material gave the hint: the app had an NPS of 84, but the subscription was around 45; 36% of users opened the app fewer than twice a month, and only 48% renewed. My read was that customers pay the fee and never see what it covers.
What the app does
- Every day: how much the subscription has covered this month and the club perk that fits where the customer is.
- Every month: the open statement, separating what was done, what is contract and what is an estimate.
- Any time: "is it worth it?", comparing against buying outright or financing, using FIPE prices and public sources.
- At the end of the cycle: a recap to share.
The number comes from the rules, the explanation from the model
The assistant uses tool use: the model never writes a figure that did not come out of the C# calculation engine. It interprets the question ("what if I paid 40% upfront?"), calls the tool and explains the result. Actions like renewing or switching cars stay in the app, always with confirmation. Without an API key, the app falls back to guided questions that call the same engine.
