A native Android application to explore US economic data. FRED Analysis pulls time series from the Federal Reserve Bank of St. Louis (FRED) API, lets the user pick any series and date range, and renders the results as an interactive line chart paired with two written summaries: a deterministic statistical readout and an AI-generated narrative from Anthropic's Claude API. It's built independently in Kotlin and Jetpack Compose to leverage production Android patterns: a Hilt-injected MVVM architecture, unidirectional StateFlow UI state, coroutine-based concurrency, and caching.
| Explore | Chart |
|---|---|
 |
 |
The app is organized around two screens connected by a bottom navigation bar:
- Explore: enter a FRED series ID (for example
GDP,GNPCA,CPIAUCSL) and a start/end date, add up to four series to compare on one chart, and save series as favorites for quick reuse. - Chart: fetches every selected series, plots them together as a line chart, shows a written insight per series (net change, period average, range, and recent direction), and generates an AI response that reads the data.
- Multi-series comparison: overlay up to four FRED series on a single chart to compare trends
- Concurrent fetching: all selected series load in parallel with coroutines (
async/awaitAll) rather than sequentially - Auto-generated insights: each series is summarized in plain language (percent change, average, min/max range, recent movement)
- AI insights: the series statistics are sent to Anthropic's Claude API, which returns a read of what the data shows; the feature is optional and the summary remains when no API key is configured or the request fails
- Persistent favorites: starred series are saved across launches via
SharedPreferences - In-memory LRU cache: a bounded, access-ordered cache avoids re-fetching identical series/date-range requests, with a manual pull-to-refresh that bypasses it
- Form validation: series ID and
YYYY-MM-DDdate checks with clear inline errors before a request is made - Loading and error states: network failures surface a friendly retry message instead of crashing
- Deep-linkable chart route: the chart is a self-contained navigation destination (
graph/{seriesIds}/{startDate}/{endDate}) whose inputs are encoded in the route and restored fromSavedStateHandle, so it survives configuration changes
| Layer | Choice |
|---|---|
| Language | Kotlin |
| UI | Jetpack Compose, Material 3 |
| Architecture | MVVM, unidirectional StateFlow state |
| Dependency injection | Hilt (Dagger) |
| Navigation | Navigation Compose |
| Networking | Retrofit, OkHttp, Gson |
| AI | Anthropic Claude API (Messages API) |
| Concurrency | Kotlin Coroutines |
| Charts | compose-charts (ehsannarmani) |
| Min / Target SDK | 28 / 36 |
The app follows MVVM with constructor injection throughout. Composable screens observe immutable UI state from a ViewModel, and all data access goes through the model layer.
com.fred_analysis/
├── MainActivity.kt # Compose host
├── MainApplication.kt # @HiltAndroidApp entry point
├── NavWrapper.kt # Scaffold, bottom nav, and NavHost routes
├── model/
│ ├── FREDApiService.kt # Retrofit interface + response models
│ ├── RetrofitInstance.kt # Retrofit/OkHttp/Gson setup (singleton)
│ ├── FavoritesRepository.kt # favorites persisted in SharedPreferences
│ └── FredObservationCache.kt# bounded LRU cache of observations
├── viewmodel/
│ ├── HomeViewModel.kt # Explore form state, validation, favorites
│ ├── GraphViewModel.kt # fetch, cache, insight generation
│ └── StartupViewModel.kt # launch state
└── ui/
├── screens/ # HomeScreen, GraphScreen, LaunchScreen
└── theme/ # Color, Type, Theme (Material 3)
State flow. Each ViewModel exposes a single immutable uiState (StateFlow); screens render it and send events back through plain function calls. GraphViewModel reads its route arguments from SavedStateHandle, so the chart survives configuration changes and process death.
Data layer. GraphViewModel requests each series concurrently, checks the FredObservationCache first, and falls back to the FRED API through RetrofitInstance. Results are cached by (seriesId, startDate, endDate) in an access-ordered LinkedHashMap capped at 12 entries.
Insights. After data loads, the ViewModel computes a short natural-language summary (direction, percent change, average, range, and recent trend) so the chart is paired with a readable takeaway.
- Open the project in Android Studio (a recent Koala/Ladybug build or newer).
- Request a free FRED API key (takes a minute) at https://fred.stlouisfed.org/docs/api/api_key.html.
- Add the key to
local.properties(git-ignored):fred.api.key=YOUR_KEY. It is exposed to the app at build time viaBuildConfig.FRED_API_KEY. - Run the
appconfiguration on an emulator or device (min SDK 28).
./gradlew assembleDebug # build a debug APK
./gradlew installDebug # install on a connected device/emulatorBuilt by Brian Hu as an independent project to learn production Android architecture. Economic data provided by the Federal Reserve Bank of St. Louis (FRED). This app is not affiliated with or endorsed by the Federal Reserve.
© 2026 Brian Hu. All rights reserved.