MyBudgetNerd: Shipping Privacy-First ML Personal Finance to the App Store

Problem

Budgeting apps typically force a bad trade: hand over your bank credentials for automatic syncing, or do everything manually and abandon the app in two weeks. Privacy-conscious users, a large and underserved group, refuse the first option and can't sustain the second.

Challenge

The product had to deliver automation-grade convenience without bank logins. That meant solving PDF statement parsing across many bank formats, transaction categorization without a cloud of labeled user data, and doing it all with an architecture that could honestly claim privacy-first.

Solution

Users import the PDF statements they already download from their banks. The pipeline extracts every transaction, an ML model categorizes them (learning from the user's corrections), and the analytics layer surfaces trends, forecasts, and statistically flagged anomalies.

AI-powered recommendations are optional and human-in-the-loop: context-aware guidance the user can turn on, not a black box making decisions. Statements are processed in memory, and users control exactly what's shared.

Technical Implementation

The product runs React on the front end with FastAPI + Python services behind it, in Docker on a dedicated EC2 instance inside its own AWS account. The categorization pipeline combines a trained classifier with correction feedback; the Oracle's forecasting and anomaly detection run against each user's own transaction history, in memory, per request.

• PDF parsing across heterogeneous bank statement formats
• ML categorization with user-correction feedback loop
• Forecasting and statistical anomaly detection per user
• In-memory processing; no bank credentials required
• Docker on EC2 in a dedicated AWS account, isolated from every other product

Results

MyBudgetNerd shipped to the App Store and operates today as a subscription SaaS with active subscribers. It holds 5.0-star reviews, with users specifically praising the clean interface, the PDF import workflow, and the privacy-conscious design. That's validation that the architecture bet was the right one.

Lessons Learned

Privacy is a product feature when it's structural. Because the architecture genuinely never touches bank credentials, the privacy story required no fine print, and reviewers noticed.

Optional AI beats mandatory AI in consumer products. Making recommendations opt-in built trust with exactly the users most skeptical of AI. Several of them said so in reviews.