Build Intelligent Android Apps with Google's AI

Jingyu Shi, from the Developer Relations Team on AI on Android, highlights the exciting innovations happening within the developer community. They're leveraging Google's AI to create applications that are creative, intelligent, and helpful in everyday life. The focus is on bringing these intelligent experiences directly to mobile devices.

Gemini: Powering Productivity, Creativity, and Expressiveness

Android is rapidly innovating with the power of Gemini to enhance user productivity, creativity, and expressiveness. This starts with assistant experiences rebuilt with Gemini, making them more powerful and helpful than ever before. Gemini can understand intent, follow trains of thought, and complete complex tasks using favorite applications.

AIPowered App Showcases

Google has showcased a range of updated applications like Recorder, TalkBack, Screenshots, and Pixel Studio, all leveraging Google's AI. They also highlighted apps like Hiker (offline hiking app), Life (personal diary/journal app), and a meal planner app, demonstrating how AI can be integrated into diverse applications.

Building with AI: A Full Tech Stack

Google provides a comprehensive tech stack to enable developers to build AI into their applications, offering an endtoend experience. This ranges from advanced AI models running in the Cloud on Google's infrastructure to the efficient Gemini Nano model, optimized for mobile devices.

Choosing Between OnDevice and Cloud Models

The choice between ondevice and cloud solutions depends on the specific use case requirements and the target market. Device reach and AI feature needs are key considerations.

Advantages of OnDevice Models (Gemini Nano)

• Processes data locally, reducing the need for data transfer.
• Functions fully offline, ensuring consistent performance.
• Lowers cost and reduces latency.

Getting Started with Gemini Nano

Experimental access to Gemini Nano has been available since October. Follow these steps to begin building:

1. Sign up for experimental access.
2. Add the Google AI Edge SDK dependency to your Gradle file.
3. Customize model parameters (temperature, topK, etc.).
4. Initiate the model with configurations.
5. Optionally, provide download configurations to handle events during model download.

Inference with Gemini Nano

Use coroutines to run `generateContent` with your prompt. For system prompts and user input, create a content object with each line of input and pass it into the `generateContent` method.

AICore: Simplifying OnDevice AI

The Android system service AICore handles the complexities of interacting with Gemini Nano, including safety features and hardware acceleration. For key use cases requiring higher quality, AICore also supports finetuning using lowrank adaptation (LoRA).

Resources for Gemini Nano

Explore the developer guide on the Android developer documentation page and check out the sample app on GitHub (available in Kotlin and Java).

Beyond Gemini Nano: Gemini API and Vertex AI for Firebase

For more complex use cases requiring information extraction and structured output (e.g., JSON), use the larger and more capable models accessible through the Gemini API in the Android SDK.

Prototyping with Google AI Studio

Google AI Studio is a great starting point for prototyping use cases.

Vertex AI for Firebase SDK

Simplify building directly in your Android app with the Vertex AI for Firebase SDK, allowing you to call the Gemini API from your Android app using languages like Kotlin, Java, or Dart, without needing a backend service. You can use more advanced models like Gemini 1.5 Flash and Gemini 1.5 Pro. Vertex AI for Firebase is now in SGA, so you can use it in production apps.

RealWorld Examples: The Life App

Jomin George shared his experience building with Gemini 1.5 Flash to create an AI assistant in the Life personal diary app. This assistant engages in conversations with users and generates personalized journal entries capturing the essence of the conversation.

AIPowered Productivity in Android Studio

Google is incorporating AI to accelerate developer productivity within Android Studio. Gemini in Android Studio helps developers build apps faster across all stages of the development lifecycle.

Gemini in Android Studio Features (Canary Channel)

• Compose Preview Generation: Easier Compose building.
• Prompt Library: Stores frequently used prompts for singleclick application.
• Gemini Code Transforms: Select code snippets and apply changes using Gemini.
• Commit Message Generation, App Quality Insights, Unit Test Scenario Generation, and more.

Conclusion

Explore the possibilities of generative AI to delight your users. Learn more about Gemini Nano, Gemini API in Vertex AI for Firebase, and the new features in Gemini Android Studio in the Android developer documentation.