Manufacturing-ready generative AI on Android with Vertex AI in Firebase

Posted by Thomas Ezan – Sr. Developer Relation Engineer (@lethargicpanda)

Gemini may also help you construct and launch new person options that can enhance engagement and create customized experiences to your customers.

The Vertex AI in Firebase SDK allows you to entry Google’s Gemini Cloud fashions (like Gemini 1.5 Flash and Gemini 1.5 Professional) and add GenAI capabilities to your Android app. It grew to become typically obtainable final October which suggests it is now prepared for manufacturing and it’s already utilized by many apps in Google Play.

Listed here are suggestions for a profitable deployment to manufacturing.

Implement App Examine to stop API abuse

When utilizing the Vertex AI in Firebase API it’s essential to implement sturdy safety measures to stop unauthorized entry and misuse.

Firebase App Examine helps defend backend sources (like Vertex AI in Firebase, Cloud Capabilities for Firebase, and even your individual customized backend) from abuse. It does this by testifying that incoming visitors is coming out of your genuine app operating on an genuine and untampered Android machine.

To get began, add Firebase to your Android undertaking and allow the Play Integrity API to your app within the Google Play console. Again within the Firebase console, go to the App Examine part of your Firebase undertaking to register your app by offering its SHA-256 fingerprint.

Then, replace your Android undertaking’s Gradle dependencies with the App Examine library for Android:

dependencies {
    // BoM for the Firebase platform
   implementation(platform("com.google.firebase:firebase-bom:33.7.0"))

    // Dependency for App Examine
    implementation("com.google.firebase:firebase-appcheck-playintegrity")
}

Lastly, in your Kotlin code, initialize App Examine earlier than utilizing every other Firebase SDK:

Firebase.initialize(context)
Firebase.appCheck.installAppCheckProviderFactory(
    PlayIntegrityAppCheckProviderFactory.getInstance(),
)

To reinforce the safety of your generative AI function, it’s best to implement and implement App Examine earlier than releasing your app to manufacturing. Moreover, in case your app makes use of different Firebase companies like Firebase Authentication, Firestore, or Cloud Capabilities, App Examine supplies an additional layer of safety for these sources as nicely.

As soon as App Examine is enforced, you’ll be capable to monitor your app’s requests within the Firebase console.

You possibly can study extra about App Examine on Android within the Firebase documentation.

Use Distant Config for server-controlled configuration

The generative AI panorama evolves rapidly. Each few months, new Gemini mannequin iterations grow to be obtainable and a few fashions are eliminated. See the Vertex AI in Firebase Gemini fashions web page for particulars.

Due to this, as an alternative of hardcoding the mannequin identify in your app, we advocate utilizing a server-controlled variable utilizing Firebase Distant Config. This lets you dynamically replace the mannequin your app makes use of with out having to deploy a brand new model of your app or require your customers to select up a brand new model.

You outline parameters that you simply wish to management (like mannequin identify) utilizing the Firebase console. Then, you add these parameters into your app, together with default “fallback” values for every parameter. Again within the Firebase console, you’ll be able to change the worth of those parameters at any time. Your app will robotically fetch the brand new worth.

This is easy methods to implement Distant Config in your app:

// Initialize the distant configuration by defining the refresh time
val remoteConfig: FirebaseRemoteConfig = Firebase.remoteConfig
val configSettings = remoteConfigSettings {
    minimumFetchIntervalInSeconds = 3600
}
remoteConfig.setConfigSettingsAsync(configSettings)

// Set default values outlined in your app sources 
remoteConfig.setDefaultsAsync(R.xml.remote_config_defaults)

// Load the mannequin identify
val modelName = remoteConfig.getString("model_name")

Learn extra about utilizing Distant Config with Vertex AI in Firebase.

Collect person suggestions to judge influence

As you roll out your AI-enabled function to manufacturing, it’s important to construct suggestions mechanisms into your product and permit customers to simply sign whether or not the AI output was useful, correct, or related. For instance, you’ll be able to incorporate interactive components reminiscent of thumb-up and thumb-down buttons and detailed suggestions types inside the person interface. The Materials Icons in Compose package deal supplies prepared to make use of icons that can assist you implement it.

You possibly can simply monitor the person interplay with these components as customized analytics occasions through the use of Google Analytics logEvent() operate:

Row {
   Button (
      onClick = {
         firebaseAnalytics.logEvent("model_response_feedback") {
            param("suggestions", "thumb_up")
         }
      }
   ) {
      Icon(Icons.Default.ThumbUp, contentDescription = "Thumb up")
   },
   Button (
      onClick = {
         firebaseAnalytics.logEvent("model_response_feedback") {
            param("suggestions", "thumb_down")
         }
      }
   ) {
      Icon(Icons.Default.ThumbDown, contentDescription = "Thumb down")
   }
}

Study extra about Google Analytics and its occasion logging capabilities within the Firebase documentation.

Consumer privateness and accountable AI

While you use Vertex AI in Firebase for inference, you’ve the assure that the info despatched to Google gained’t be utilized by Google to coach AI fashions (see Vertex AI documentation for particulars).

It is also essential to be clear along with your customers once they’re partaking with generative AI know-how. It’s best to spotlight the potential of surprising mannequin conduct.

Lastly, customers ought to have management inside your app over how their exercise associated to AI mannequin interactions is saved and deleted.

You possibly can study extra about how Google is approaching Generative AI responsibly within the Google Cloud documentation.