Birikim
A personal finance assistant that goes past tracking and tells you what to do next.
- iOS
- Android
- 2026
- My own product
Screenshots
Click an image to enlarge it.
About the app
Most budgeting apps on the market show you how much you spent last month. I wrote Birikim because that is not the question that actually helps: people do not want the past, they want to know what to do right now.
So I put forecasting at the centre of the app. Using past data and your weekly spending rhythm, it works out where you will land at the end of the month at this pace. It also compares category by category, showing which line item has drifted outside your own average.
The part I spent the most time on was Turkish natural language input. When someone types "35 tl kahve içtim" ("spent 35 lira on coffee"), the amount, the category and the date all have to be extracted, and Turkish suffixes plus free-form spelling make this noticeably harder than English. For itemised receipt parsing I run the OCR on the device — the receipt image never leaves the phone.
The subscription detector picks up recurring payments and questions which ones you actually use. The investment simulator makes the "what would you have today if you had saved 500 ₺ a month" question concrete; it is built as an educational module that points people to licensed institutions.
The app is under active development. The MVP scaffold and the UI layer are ready; the Firebase integration, the AI service connection and the store release are in the next phases.
Key features
- Predictive budget: end-of-month forecast from past data and weekly rhythm
- Category insights: which line item has drifted outside your own average
- AI coach: a weekly chat that reviews your financial habits
- Natural language input: an expense logged from a free-form Turkish sentence
- Receipt scanning: itemised parsing from a photo, with on-device OCR
- Subscription detector: finds recurring payments and flags the unused ones
- Investment simulator: an educational module showing what regular saving would be worth today
- Alternative suggestions: grocery and subscription comparisons specific to the Turkish market
Technologies used
51 technologies · 9 categoriesLanguages and Platform
- Dart
- Flutter 3.44
- One codebase, two platforms
- iOS and Android targets
- Custom theme on Material 3
- Handling platform differences
Architecture and State Management
- Feature-first architecture
- Clean Architecture layers
- data / domain / presentation separation
- Riverpod 2
- Code generation
- Type-safe routing with GoRouter
- Testable service layer
Local Data
- Drift (type-safe layer over SQLite)
- Schema and migration management
- Offline operation
- Local cache
- Query optimisation
Backend
- Firebase Authentication
- Cloud Firestore
- Firebase Storage
- Firebase Analytics
- Crashlytics
- Firebase Cloud Messaging
- Security rules
AI and Text Processing
- Anthropic Claude Haiku 4.5
- Parsing expenses from natural language
- Turkish suffixes and spelling variations
- Prompt design for the coaching chat
- Token cost control
- Response validation and fallback
On-Device Processing
- Google ML Kit
- OCR from receipt photos
- Text region parsing
- Images never leave the device
- Camera and gallery permissions
UI and Visualization
- Charts with fl_chart
- Design system tokens (colour, typography, spacing, radius)
- Manrope typography
- Full light/dark theming
- Empty and loading states
- Accessible contrast
Monetization
- RevenueCat
- Freemium subscription
- Free / premium entitlement split
- Subscription status verification
- Affiliate referrals
Localization
- Turkish (primary)
- English infrastructure in place
- Currency and date formatting
- Turkish characters and collation
Idea, design, UI, code, store release and post-launch maintenance — I ran the entire process end to end on my own. No team, no agency, no outsourcing.
Development notes
- Languages and approachiOS and Android from a single codebase with Flutter and Dart. I set the architecture up as feature-first + Clean Architecture; every feature has its own data, domain and presentation layers.
- StatusUnder active development. The MVP scaffold and the UI layer are ready; the Firebase integration, the AI service connection and the store release are the next phases.
- The hard partTurkish natural language input. Extracting an amount, a category and a date from a freely written sentence is noticeably harder than in English because of Turkish suffixes.
- Privacy decisionReceipt scanning runs on the device; the receipt image never leaves the phone. It also lowers cost, but privacy was the real reason.
- Product ownershipThe idea, the design, the architecture and all of the code are mine.
Let's talk about Birikim
I'm happy to walk you through the technical details, the problems I ran into and how I solved them.