From AI Prototype to a Production App

Building an AI feature is just the beginning; scaling it into a production-ready application is where the real challenge begins. Using a meal-planning app, you will see how to bridge the gap between a great idea and a polished product using Google’s Gemini models, Firebase AI Logic, and Remote Config. This session will cover:

- How to analyze images of a fridge, cupboard or grocery store shelf, identify ingredients and generate meal recommendations.
- How to create mouth-watering visuals with different AI models based on generated recipes.
- How to keep prompts secure in the backend and use Remote Config to deliver new prompt versions without requiring an app update or affecting the current experience.

Whether you are still on the prototyping phase, or trying to figure out how to manage AI lifecycle in a live app, this session will provide you with a roadmap to move from experimental code to a reliable production environment.

Transcript

1. From AI Prototype to a Production App Developer Relations Engineer Marina Coelho
3. 3 Extract ingredients from a photo Generate recipe based on ingredients list Generate recipe image based on recipe title
4. 4 Store prompts securely and remotely Store images and recipes securely and remotely Filter recipes list based on various criteria
5. 5 Prototyping
8. 8 Building Friendly Meals
10. BUILD RUN A/B Testing Crashlytics App Distribution Remote Config Test Lab Cloud Messaging Performance Monitoring Google Analytics Cloud Functions for Firebase Cloud Storage for Firebase Data Connect database App Hosting Authentication Realtime Database Firestore database Hosting App Check Firebase AI Logic
11. 11 Firebase integrates with many Google products to bring the best app development experiences to developers
12. Feature Multimodal input Structured output Media output Prompt storage Use case Upload image and detect ingredients Generate recipe based on list of ingredients and return it in a json format Store prompts remotely Generate recipe image Store recipe images remotely File storage Store recipes remotely Remote database Filter recipes list based on various criteria Powerful search queries Firebase Firebase AI Logic Firebase AI Logic Firebase AI Logic Firebase AI Logic Cloud Storage Firestore Firestore 🔒 Authentication
13. 13 Firebase AI Logic gives you access to the latest generative AI models from Google: the Gemini models and Imagen models. If you need to call the Gemini API or Imagen API directly from your mobile or web app, rather than server-side, you can use the Firebase AI Logic client SDKs.
14. 14 Server Prompt Templates • Create a Server Prompt Template • Provide a key (template ID) in your app's codebase • Update your prompt and configuration without releasing a new app version • Protect against exposing your prompt client-side • Supported with Gemini and Imagen models
15. 15 Prompts and model configuration can change, and you may need to create new versions of your template at some point. You can use Firebase Remote Config to deliver the latest template ID to app users without requiring an app update or affecting their current experience.
16. 16 Setting up your environment
17. 1. Create a Firebase project 2. Configure services in the Firebase console 3. Add Firebase dependencies to the build.gradle file
18. dependencies { // ... other dependencies implementation(platform("com.google.firebase:firebase-bom:34.8.0")) implementation("com.google.firebase:firebase-ai") implementation("com.google.firebase:firebase-config") implementation("com.google.firebase:firebase-storage") implementation("com.google.firebase:firebase-firestore") implementation("com.google.firebase:firebase-auth") }
19. 19 Creating a model instance
20. @Provides fun firebaseAI(): FirebaseAI { return Firebase.ai(backend = GenerativeBackend.googleAI()) } @OptIn(PublicPreviewAPI::class) @Provides fun generativeModel(ai: FirebaseAI): TemplateGenerativeModel { return ai.templateGenerativeModel() }
21. @Provides fun firebaseAI(): FirebaseAI { return Firebase.ai(backend = GenerativeBackend.googleAI()) } @OptIn(PublicPreviewAPI::class) @Provides fun generativeModel(ai: FirebaseAI): TemplateGenerativeModel { return ai.templateGenerativeModel() }
22. 22 Extracting ingredients from a photo
24. suspend fun generateIngredients(imageData: String): String { val response = generativeModel.generateContent( templateId = remoteConfig.getString(GENERATE_INGREDIENTS_KEY), inputs = mapOf( MIME_TYPE_FIELD to MIME_TYPE_VALUE, IMAGE_DATA_FIELD to imageData ) ) return response.text.orEmpty() }
25. 25 Generating recipe from ingredients list
27. @Serializable data class RecipeSchema( val title: String = "", val instructions: String = "", val ingredients: List<String> = listOf(), val prepTime: String = "", val cookTime: String = "", val servings: String = "", val tags: List<String> = listOf() )
28. suspend fun generateRecipe(ingredients: String, notes: String): RecipeSchema? { val response = generativeModel.generateContent( templateId = remoteConfig.getString(GENERATE_RECIPE_KEY), inputs = buildMap { put(INGREDIENTS_FIELD, ingredients) if (notes.isNotBlank()) { put(NOTES_FIELD, notes) } } ) return response.text?.let { json.decodeFromString<RecipeSchema>(it) } }
29. 29 Generating an image from recipe title
31. suspend fun generateRecipePhoto(recipeTitle: String): Bitmap? { val response = generativeModel.generateContent( templateId = remoteConfig.getString(GENERATE_RECIPE_PHOTO_GEMINI_KEY), inputs = mapOf(RECIPE_TITLE_FIELD to recipeTitle) ) return response.candidates.firstOrNull()?.content?.parts ?.filterIsInstance<ImagePart>()?.firstOrNull()?.image }
32. 32 Storing recipes securely and remotely
33. suspend fun createAnonymousAccount(): User { val authResult = auth.signInAnonymously().await() return User(authResult.user?.uid.orEmpty()) }
34. suspend fun addImage(image: Bitmap): String { val scaledBitmap = scaleBitmap(image) val stream = ByteArrayOutputStream() scaledBitmap.compress(Bitmap.CompressFormat.WEBP, 70, stream) val data = stream.toByteArray() val randomId = UUID.randomUUID().toString() val imagesRef = storageRef.child("images/$randomId.webp") imagesRef.putBytes(data).await() return imagesRef.downloadUrl.await().toString() }
35. suspend fun addRecipe(recipe: Recipe): String { val recipeRef = firestore.collection(RECIPES_COLLECTION).add(recipe).await() return recipeRef.id } suspend fun getRecipe(recipeId: String): Recipe { val recipePath = "${RECIPES_COLLECTION}/${recipeId}" return firestore .pipeline() .documents(recipePath) .execute().await().results.toRecipe() }
36. suspend fun addRecipe(recipe: Recipe): String { val recipeRef = firestore.collection(RECIPES_COLLECTION).add(recipe).await() return recipeRef.id } suspend fun getRecipe(recipeId: String): Recipe { val recipePath = "${RECIPES_COLLECTION}/${recipeId}" return firestore .pipeline() .documents(recipePath) .execute().await().results.toRecipe() }
37. 37 Filtering recipes list
38. Let's dive into Pipelines!
39. 39 Let’s run the app!
42. 42 Going to production
43. App Check Protect your backend
44. dependencies { // ... other dependencies implementation(platform("com.google.firebase:firebase-bom:34.8.0")) implementation("com.google.firebase:firebase-ai") implementation("com.google.firebase:firebase-config") implementation("com.google.firebase:firebase-storage") implementation("com.google.firebase:firebase-firestore") implementation("com.google.firebase:firebase-auth") // strongly recommended implementation("com.google.firebase:firebase-appcheck-playintegrity") }
45. class MainActivity : ComponentActivity() { override fun onCreate(savedInstanceState: Bundle?) { super.onCreate(savedInstanceState) Firebase.initialize(context = this) Firebase.appCheck.installAppCheckProviderFactory( PlayIntegrityAppCheckProviderFactory.getInstance(), ) } }
48. 48 Challenge time!
49. 49 Danke! What’s next: • firebase.google.com/docs • bit.ly/FriendlyMealsAndroid Developer Relations Engineer Marina Coelho
50. Appendix
51. 51 Cloud Storage for Firebase is a powerful, simple, and cost-effective object storage service. You can use our client SDKs to store images, audio, video, or other user-generated content.
52. 52 Firestore is a flexible and scalable NoSQL database for mobile, web, and server development. It keeps your data in sync across client apps through realtime listeners and offers offline support for mobile and web so you can build responsive apps that work regardless of network latency or Internet connectivity.
54. 54 Firebase Authentication provides backend services, easy-to-use SDKs, and ready-made UI libraries to authenticate users to your app. It supports authentication using passwords, phone numbers, popular federated identity providers like Google, Facebook and Twitter, and more.
56. 56 Firebase App Check helps protect your app backends from abuse by preventing unauthorized clients from accessing your backend resources. It works with both Google services (including Firebase and Google Cloud services) and your own custom backends to keep your resources safe.