Inspiration
For over 4 to 5 years, I was a dedicated user of apps like Money Manager, manually logging every single coffee, grocery run, and bill payment by hand.
While manual tracking gave me control, it was exhausting. On busy days, receipts piled up, transactions got forgotten, and tracking fatigue inevitably set in. But when I looked for modern automated alternatives, I was horrified by what I found: almost every automated expense tracker forced you to either link your net-banking credentials or upload your private bank SMS messages to remote cloud servers to be processed and monetized.
I refused to trade my privacy for convenience. I asked myself: Why can't we have the effortless automation of automatic tracking with the absolute privacy and local control of classic Money Manager?
That frustration sparked the obsession behind Smart Money AI: build an intelligent, modern expense tracker that automatically categorizes bank transactions the moment an SMS arrives—powered by an embedded, on-device AI model that runs in <15ms with 100% privacy and zero data leaving the phone.
What it does
Smart Money AI eliminates manual bookkeeping while keeping your personal finances completely private:
- Sub-15ms On-Device AI Classification: Powered by a fine-tuned, quantized TinyBERT transformer model running locally via ONNX Runtime. It detects amounts, merchants, categories, and account types in real-time.
- 100% Offline & Private: No accounts required, no telemetry, and no servers. Your financial history belongs to you alone.
- Global Multi-Currency & Regional Support: Automatic country detection, international currency formatting, and dynamic regex support for worldwide financial institutions.
- Modern Budgeting & Insights: Interactive visual analytics, custom category creation, transfer detection, and smart duplicate filtering.
- Encrypted Backups: Optional client-side encrypted backup directly to the user’s personal Google Drive.
- RevenueCat Pro Access: Dual-tier monetization unlocking Pro features (unlimited custom categories, deep analytics, cloud backups) via Google Play Billing synced with RevenueCat.
How we built it
- Modern Android Stack: Built from scratch with 100% Kotlin, Jetpack Compose, and Material 3 design for fluid 60fps micro-animations and seamless dark-mode ergonomics.
- Embedded Machine Learning: Fine-tuned TinyBERT model quantized to INT8 (<15MB footprint), running on ONNX Runtime and engineered with 16KB memory page-alignment for next-gen Android 15 compatibility.
- RevenueCat Purchases SDK v10+: Integrated using
PurchasesAreCompletedBy.MY_APP. This architectural decision lets our app maintain native Google PlayBillingClientautonomy for checkout while RevenueCat automatically observes transactions, validates receipts server-side, and synchronizes entitlements in real time. - Local Architecture: Offline-first architecture using Room Database, Kotlin Coroutines, and reactive Kotlin StateFlows.
Challenges we ran into
- Ultra-Low Latency Inference: Running transformer models on mobile often leads to sluggish UI or battery drain. We spent weeks quantizing and tuning our TinyBERT model to execute in under 15ms without heating or battery degradation.
- Global SMS Diversity: Bank SMS formats vary wildly worldwide. We built a hybrid classification engine combining regex token pre-parsing with machine-learning contextual classification to handle ambiguous merchant strings.
- Seamless Monetization: Ensuring that users who redeem Google Play promo codes or purchase lifetime passes have their entitlements validated instantly by RevenueCat without disrupting existing local storage states.
Accomplishments that we're proud of
- Zero Cloud Leaks: Complete privacy guarantee. All NLP parsing happens on-device.
- Post-Release Organic Momentum: First shipped on Google Play on August 30, 2026, and rapidly scaled to 50+ organic active users within weeks with zero paid advertising!
- Robust RevenueCat Integration: Full production integration verified with live Google Play receipt synchronization, active trials, and sandbox testing.
What we learned
Building a financial tool in public taught us that transparency is the ultimate differentiator. Users are fatigued by invasive apps that monetize their personal habits. When people see that AI can run locally in <15ms without sending data to a server, their trust and engagement skyrocket.
What's next for Smart Money AI
We are building the future of private, autonomous personal finance. Here is what is on our immediate product roadmap:
On-Device AI Personalization (Upcoming Pro Feature): Unlike any other finance app on the market, we are introducing deep personalization based on individual transaction history and spending rhythms—executed 100% on-device without storing or transmitting personal data. Your financial assistant learns your lifestyle entirely locally.
Predictive Goal & Investment Forecasting: Smart financial analytics that model future cash runway, recurring liabilities, and predict timelines for achieving user investment and savings goals.
Smart Cognitive AI Budgeting: Dynamic budget limits that automatically adjust based on seasonality, unexpected emergency expenses, and changing cash-flow habits.
Smart Bank Statement Auto-Import: Zero-cloud, client-side parser to ingest historical bank statements (PDF / CSV) into your ledger with automatic merchant deduplication.
Adaptive Self-Learning Categorization: An on-device learning loop that adapts to user re-categorizations and custom preferences, continually improving local model accuracy over time.
Conversational Voice Input: Natural-language voice logging (e.g., "Spent $24 on dinner with Alex") with on-device speech-to-text and instant entity extraction.
AI-Powered Smart Digest: A weekly on-device summary providing actionable spending diagnostics, trend alerts, and proactive tips to optimize savings.
Built With
- android
- android-15
- coroutines
- deep-learning
- fintech
- flows
- google-drive-api
- google-play-billing
- huggingface
- jetpack-compose
- kotlin
- machine-learning
- material-design-3
- natural-language-processing
- offline-first
- onnx-runtime
- personal-finance
- privacy
- python
- pytorch
- revenuecat
- room-database
- tinybert
- transformers
Log in or sign up for Devpost to join the conversation.