MyBudgetNerd is a consumer personal-finance product we took from concept to the App Store, now a subscription SaaS with active subscribers. This case study covers the privacy-first architecture, the ML pipeline behind it, and what shipping a consumer product end-to-end demonstrates about our delivery capability.
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 transactions for review, 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. Those reviews are qualitative product feedback, separate from measured financial outcomes.
Lessons Learned
Privacy choices belong in the workflow. Importing statements avoids bank credentials; optional AI has its own consent boundary. Those are specific behaviors a user can evaluate.
Optional AI makes the choice meaningful. Core budgeting remains available without it, so a user can decline recommendations and still use the product.
Explore further
MyBudgetNerd
Import a bank statement, review categorized transactions, and catch unusual charges—no bank login needed.
The Power of Privacy: How Ephemeral Backends Make Our Products Harder to Breach
How Aegis BI and MyBudgetNerd minimize stored data, why Thera needs controlled persistence, and what those choices mean for privacy and security.
Practical Forecasting for Small-Business Finance
You don't need a data science team to forecast cash and revenue. You need clean history, honest uncertainty, and models you can explain to whoever bets on them.
Rebuilding the Infrastructure: One Cluttered AWS Account Into Four
Two live products shared one AWS account, three deployment systems, and a single disk. Over four days we rebuilt it into an account per app without taking either product down, plus the two near-misses that came closest to making that sentence untrue.
Have a Project in Mind?
Tell us what needs to work better. We’ll help define the first step.