How I "Vibe Coded" a Secure AI Finance App (Without Writing the Boilerplate)

Swofski — Mon, 24 Nov 2025 11:31:17 +0000

Introduction

I have a degree in computing, but my day job is booking water efficiency appointments. I wanted to build a serious finance tool to fix my own spending, but I didn't want to spend six months writing HTML tags and CSS manually.

I decided to act less like a "coder" and more like an Architect. I used Google Gemini as my "Junior Developer" to build Finance Buddy, a PWA with bank-grade encryption.

Here is how I orchestrated an AI to build a production-ready app, and why "Vibe Coding" is the future for solo founders.

1. The Architecture: Privacy by Design

The biggest challenge wasn't the code; it was the security model. I didn't want to see user data, and I didn't want the AI to leak it either. I defined a "Zero-Knowledge" architecture:

Client-Side: The browser encrypts the data (AES-GCM) using the user's password.
The Transfer: Only encrypted "blobs" are sent to Firebase.
The AI: The Gemini AI operates on a stateless, zero-retention basis. We only send anonymized, ephemeral snippets of data to the inference engine.
The AI processes the request and forgets it immediately, ensuring your financial history is never used to train external models.

My role wasn't typing the encryption functions; it was verifying that the AI implemented the security protocols I designed.

2. Managing the "Junior Dev" (Gemini)

Most people try to get AI to "write the whole app." That fails. I treated Gemini like a junior teammate.

Bad Prompt: "Make a finance app."
My Strategy: "Write a React hook that takes a transaction object and encrypts the amount field using the Web Crypto API. Return the error state if encryption fails."

I spent my time reviewing code, not writing it. This allowed me to move 10x faster than a traditional dev. I focused on the features (Receipt Scanning, Forecasting) rather than missing semicolons.

3. The Result: From Concept to "Shipped"

The result is a fully functional PWA with:

Receipt Scanning: Integrating Gemini Vision to read paper receipts.
Interactive AI: A chat interface that knows your budget but respects your privacy.
Real Usage: It's currently live, helping me (and my first few users) track debts and subscriptions.

Conclusion

You don't need to be a 10x Engineer to build great software anymore; you need to be a 10x Architect. By leveraging AI for the heavy lifting, I built a tool that solves a real problem without quitting my day job (yet).