🌱 PlantWater

An Android app that identifies a plant species from a photo using AI, then sets up a watering schedule and sends reminder notifications automatically.

Take a photo of a plant and the Gemini API fills in its common name, scientific name, and recommended watering interval. From there, WorkManager sends periodic reminders so you never forget to water it, with a "watered" action right on the notification. Built entirely in Jetpack Compose, with Room for offline storage, WorkManager for background reminders, and CameraX for capture.

Solo project — I designed, built, and debugged the entire app end-to-end on a physical device.

---

✨ Features

Feature	Description
📷 Register a plant by photo	Capture with CameraX or pick from the gallery → the Gemini Vision API automatically fills in species, scientific name, and watering interval
🔁 Fallback on failed recognition	If the API fails or misidentifies the plant, instantly falls back to a 6-entry manual catalog — a deliberate design choice given the instability of the free-tier API
⏰ Smart reminders	WorkManager PeriodicWorkRequest schedules repeating per-plant reminders; "Watered" / "Later" can be handled right from the notification without opening the app
📅 Watering calendar	Visualizes watering history on a monthly calendar, showing days elapsed since the last watering
🗑 Edit / delete schedules	Edit plant info or reminder time; deleting a plant also cancels its scheduled reminders

📱 Screen Flow

flowchart TD
    Home["Home (plant list)"] -->|"+"| Camera["Take photo"]
    Camera -->|capture/select| Loading["AI recognition"]
    Loading -->|Gemini response| Info["Review info"]
    Info -->|register| Schedule["Create / edit schedule"]
    Schedule -->|save| Home

    Home -->|tap card| Calendar["Watering calendar"]
    Calendar -->|settings| Schedule
    Schedule -->|save| Calendar

    Notif["System notification"] -->|"Watered"| Log[(Watering log)]
    Notif -->|tap| Calendar

Loading

🛠 Tech Stack

• Language / UI: Kotlin, Jetpack Compose, Material 3
• Async: Kotlin Coroutines, Flow
• Local storage: Room (KSP)
• Background work: WorkManager (CoroutineWorker, PeriodicWorkRequest)
• Camera: CameraX (Preview, ImageCapture) + Photo Picker
• AI recognition: Google Gemini API (gemini-flash-latest, Vision + structured JSON response), implemented directly over HttpURLConnection to avoid an extra SDK dependency
• Build: Gradle Kotlin DSL, AGP, KSP

🏗 Architecture

app/src/main/java/com/moonkata/plantwater/
├── data/local/        # Room: Entities (Plant, WateringLog), Dao, Database, Repository
├── recognition/        # Gemini API client, recognition result model, manual fallback catalog
├── reminder/           # WorkManager scheduler, reminder Worker, action BroadcastReceiver
├── ui/
│   ├── home/            # Plant list
│   ├── camera/          # Capture / gallery selection
│   ├── recognition/     # AI recognition in-progress screen
│   ├── info/             # Review recognition result / manual selection
│   ├── schedule/        # Create & edit schedule (shared screen for both)
│   ├── calendar/        # Watering calendar
│   └── theme/
└── util/                 # Shared utilities (e.g. image downsampling)

Room access is centralized behind a repository, and each screen subscribes to a Flow via collectAsState so database changes are reflected in the UI immediately. Navigation is implemented directly with a sealed interface Screen + when, without a navigation library.

🔍 Problems solved along the way

Issues hit during real-device testing and live API integration (full write-up in 01_plan/progress_log.md):

• Compose state-reset bug: declared mutableStateOf at the top level of setContent {} without remember, so state reset to its initial value on every recomposition and screens immediately snapped back. Fixed by wrapping it in remember { mutableStateOf(...) } — a hands-on lesson in how Compose preserves state across recomposition.
• KSP × AGP build conflict: AGP's built-in Kotlin support conflicted with how KSP uses the kotlin.sourceSets DSL (google/ksp#2729). Worked around it with a flag in gradle.properties.
• Gemini model deprecation: the pinned model version used during development started returning 404 on a newly issued API key → switched to the always-current alias (gemini-flash-latest) for stability.
• Network timeout tuning: on networks where the connection couldn't be established at all, the default 15s timeout was retried multiple times, taking almost 2 minutes to reach the manual fallback. Lowered connect/read timeouts to 6s/12s to detect failure faster and fall back to manual entry sooner.
• Designing around API instability: treated the free-tier API's failure modes as a first-class requirement from the start, making the automatic fallback to the manual-selection dropdown a required feature rather than an afterthought.

🚀 Building

git clone https://github.com/katalog/Android-Plant-Water.git
cd Android-Plant-Water

1. Get a Gemini API key from Google AI Studio
2. Create local.properties in the project root and add:

   GEMINI_API_KEY=your_api_key_here
   

3. Open in Android Studio and run, or:

   ./gradlew assembleDebug

The app still runs without an API key — recognition simply fails and falls back to manual selection.

📌 Status

The full core flow (capture → AI recognition → register → reminder → calendar) is implemented and tested on a physical device. In progress:

• Final on-device verification of the gallery-selection flow
• On-device TFLite recognition fallback (stretch goal)
• Home screen widget (Glance)

---

Detailed screen design in 01_plan/plant_app_navigation.md, dev log in 01_plan/progress_log.md.