Aller au contenu principal
NUKOE

Build Android App with Google Generative AI: Step-by-Step Tutorial

• 10 min •
Illustration d'une application Android intégrant l'IA générative de Google.

Building an Android app that integrates generative AI is no longer reserved for teams with a research department. Google has multiplied the tools so that any Android developer can add intelligent capabilities to their app without being a machine learning expert. This tutorial guides you step by step in creating an Android app using Google Vertex AI Agent Builder, starting from a blank project to a first natural language interaction.

Why Generative AI is a Game Changer for Android Developers

The rise of language models has transformed how we design user interfaces. Instead of rigid buttons and forms, users can now converse with the app. According to a Google Cloud article, "vibe coding" — describing what you want to build in natural language — is becoming a full-fledged development method (source: Cloud Google - Vibe Coding Explained). For Android, this opens the door to integrated personal assistants, content generators, or real-time analysis tools.

The official "AI on Android" guide (source: Developer Android - AI on Android) lists available APIs: ML Kit, TensorFlow Lite, and the new APIs via Google AI Edge. But for advanced generative processing, Vertex AI Agent Builder is the most comprehensive tool.

Concrete Use Case: A Smart Task Management App

Imagine an app "SmartTasks" that allows users to add tasks, organize them, and get contextual suggestions (e.g., "What is my priority today?"). Generative AI is used to analyze tasks and propose a priority order, summarize the day, or even generate subtasks.

This case is deliberately simple to stay focused on integration. You can adapt it to your needs.

Prerequisites

  • Android Studio (latest stable version)
  • A Firebase project (free) for authentication and cloud storage
  • A Google Cloud project with billing enabled (Vertex AI is a paid service, but with a startup credit)
  • Basic knowledge of Kotlin and Android

Step 1: Create the Android Project and Add Firebase

Open Android Studio and create a new project with "Empty Views Activity". Name it "SmartTasks".

Then, add Firebase to your project. The official documentation (source: Firebase Google - Add Firebase to your Android project) details the procedure:

  1. Go to the Firebase console, create a project if not already done.
  2. Add an Android app with your project's package name.
  3. Download the `google-services.json` file and place it in your project's `app/` folder.
  4. Add Firebase dependencies in `build.gradle` (app level):
implementation(platform("com.google.firebase:firebase-bom:32.0.0"))
implementation("com.google.firebase:firebase-auth-ktx")
implementation("com.google.firebase:firebase-firestore-ktx")
  1. Sync the project.

Firebase will serve for authentication (required to secure access to Vertex AI) and for storing tasks.

Step 2: Enable Vertex AI Agent Builder

Vertex AI Agent Builder allows creating conversational agents based on language models. The official codelab (source: Codelabs Developers Google - Building AI Agents with Vertex AI Agent Builder) describes the process in detail.

  1. In the Google Cloud console, enable the Vertex AI API.
  2. Create an "Agent" in Vertex AI Agent Builder:
  • Give it a name: "SmartTasksAgent"
  • Set the base model (e.g., `gemini-1.5-pro`)
  • Add system instructions: "You are a productivity assistant. You help the user organize their tasks. You respond in French."
  1. Note the `agent_id` (looks like `projects/.../locations/.../agents/...`).
  2. Generate an API key (or use a service account) to call the agent from the Android app.

> Tip: To secure access, do not embed the key directly in the app. Use Firebase Functions as a proxy, or the Vertex AI SDK for Android (in preview).

Step 3: Integrate the Vertex AI SDK into the App

Google recently released an Android SDK for Vertex AI (in alpha). Adding it to your project is simple:

implementation("com.google.cloud:google-cloud-vertexai:0.3.0") // Check version

If you prefer a REST approach, you can use Retrofit to call the API directly. But the SDK handles authentication and streaming more easily.

Create a repository to centralize calls:

class SmartTasksRepository(private val agentId: String) {
    private val vertexAI = VertexAI.init(/ context /)

    suspend fun askAgent(prompt: String): String {
        val agent = vertexAI.agent(agentId)
        val response = agent.sendMessage(prompt)
        return response.text
    }
}

> Note: The SDK is in alpha, so the API may change. Refer to the official documentation for the latest updates.

Step 4: User Interface for Chat

Create a simple screen with RecyclerView to display messages and an input field. Use ViewModel to manage state.

class ChatViewModel(private val repository: SmartTasksRepository) : ViewModel() {
    private val _messages = MutableLiveData<List<Message>>()
    val messages: LiveData<List<Message>> = _messages

    fun sendMessage(text: String) {
        viewModelScope.launch {
            _messages.value = _messages.value + Message(text, isUser = true)
            val response = repository.askAgent(text)
            _messages.value = _messages.value + Message(response, isUser = false)
        }
    }
}

Step 5: Link Firebase Tasks to the Agent

For the agent to know the user's tasks, you need to provide context. When the user asks a question, fetch tasks from Firestore and send them as context in the prompt.

Example:

val tasks = firestore.collection("users").document(userId).collection("tasks").get().await()
val context = "Here are my current tasks: ${tasks.documents.joinToString { it.data }}"
val prompt = "$context

${userInput}"
val response = repository.askAgent(prompt)

Deployment and Testing

Test the app on an emulator or physical device. Ensure the agent responds correctly. For example:

  • User: "What is my priority today?"
  • Agent: "Based on your tasks, the priority is to finish the project report, as it is due tomorrow. Then, you have a meeting at 2 PM."

Optimization and Production Deployment

  • Error handling: Add timeouts and error messages if the agent does not respond.
  • Quota limits: Vertex AI has request-per-minute limits. Use client-side caching for frequent questions.
  • Security: Validate user input to prevent prompt injections.
  • Costs: Monitor consumption via the Google Cloud console and optimize token count by limiting context.

Generative AI on Android is still young, but Google's tools make it accessible. By combining Firebase for the backend and Vertex AI for intelligence, you can create apps that understand your users' natural language. This tutorial gave you the basics; it's up to you to imagine the possibilities.

To Go Further

These sources will allow you to deepen each step.