CalmLedger: Building Privacy-First Finance Tracking with On-Device AI

Challenge

Traditional finance apps relied on server-side processing for transaction categorization and anomaly detection. CalmLedger needed strict privacy guarantees and low-latency feedback without cloud round-trips.

Solution

We designed a local-first architecture that routes categorization and prediction through Core ML models directly on-device, with fallback heuristics for unsupported environments and explicit privacy boundaries.

Before: App -> Cloud API -> Category Result
After: App -> On-device Model -> Category Result (no network)

Implementation

• - Migrated classification logic into an actor-isolated inference module.
• - Added deterministic fixture tests for model and fallback behavior.
• - Implemented privacy-safe telemetry that excludes transaction content.
• - Added runtime profile switching for thermal-safe processing.

actor TransactionClassifier {
  func classify(_ input: TransactionInput) async throws -> Category {
    // On-device inference only
  }
}

Results

• - App Store rating: 4.8 stars after release cycle.
• - Downloads: 10,000+ cumulative installs.
• - Privacy incidents: zero cloud exposure concerns reported.

Client Testimonial

"3NSOFTS translated strict privacy requirements into a product architecture we could trust. We launched quickly without sacrificing user confidence."

Technologies Used

• - Swift 6.0, SwiftUI
• - Core ML 8.0
• - SwiftData, CloudKit

Timeline: 5 weeks