From Pigeonholes to Personalization
Today we want to write about a new development mode for our AI finance app called LedgerGenius.
In talking about what MVP features we should put into the app, we realized that we cannot possibly cover ALL of the essential use cases that each individual or family might want for their unique financial circumstances and goals.
Asking ourselves what our own financial goals are and what the purpose of using this app is, we realized that it is highly unique to our own situation. Even as creators of this app, we felt dissatisfied with the UI and UX we forced ourselves to use.
It made us realize that we need to make a big change in how we build this app.
The Discovery: Better Questions, Better Context
We first started by asking: Can we ask better questions to the user to understand their financial goal and circumstances that they are in and start from there?
We simulated this using our coding tools, giving it a prompt where we imagined an "Interviewer" personal finance manager asking 10 questions to gain a holistic understanding of the user's financial situation and goals to provide better insights.
Surprisingly, it asked very relevant questions—even more so as the interview went on, because it started asking pertinent follow-ups based on our answers.
By the end of the interview, we were getting to deep nuances, like which internal transfers should NOT count as income or expense—a detail that is notoriously hard to handle in most apps given the complexities of everyone's financial situations. We were also able to customize the goal to be very pertinent to our situation: runway calculation and monthly expense management.
One Task Remaining: On-the-Fly UI Generation
At the end, one task remained: on-the-fly UI generation.
In building DashboardGenius, we experienced that Nano Banana with Gemini 3.0 Pro does an exceptional job of generating high-fidelity visualizations that are very custom to the content.
We believe this could apply to LedgerGenius as a way to create very customized, pertinent, but also accurate visualizations of the user's financial status and progress towards their goals. Crucially, this can be done without having to engineer different UIs that ultimately "pigeonhole" users into certain buckets or categories.
The Revolution
In realizing this, we had a revolution! This could be the new way of building software:
- Build a conduit through which we thoroughly understand the user's situation via adaptive interviews.
- Integrate the data.
- Use an appropriate model to generate the outcome specifically for that user, removing the bottleneck of constrained, pre-defined UIs and UX.
