Logging an expense takes longer than making it, so people stop. Voice fixes that, but sending audio of somebody's spending to a server is exactly the trade nobody should have to make.
Case studies
Saldy — a finance app you talk to
On-device speech, private sync, no server to trust
The problem
What I did
Speech recognition on the phone
The recognition runs on the device, so no audio ever leaves it. It is more work than calling an API, and it is the only reason the privacy claim is true rather than decorative.
Sync with nothing in the middle
SwiftData with a private CloudKit database. The data lives in the user's own iCloud: I cannot read it, there is no backend to pay for, and there is nothing to breach.
Subscriptions that restore
StoreKit, including the restore path. That path is where small apps fail review, because it is the one nobody tests.
Eight languages kept honest
English is the base and the rest are translations, checked by a script that fails the build if a key is missing anywhere. Half-translated apps are shipped by people who check by hand.
Where it ended up
On the App Store, built and shipped by one person: interface, code, icon, screenshots, store listing and every round of review.
Built with: Swift · SwiftUI · SwiftData · CloudKit · StoreKit · Speech
Want something like this?
Two lines about the project is enough to start. A person reads it and a person answers.