Mobile AI Assistant Development Based on Gemini (Google)

AI Assistant Development in Mobile Application Based on Gemini (Google)

Gemini — only top model with native Android SDK from Google. google-ai-android (Generative AI SDK) integrates via Gradle without server proxy, simplifying start. But for production direct mobile access — mistake: API key extractable from APK. This contradiction between convenience and security needs solving at project start.

Google AI SDK: Android and iOS

On Android official path:

// build.gradle.kts
implementation("com.google.ai.client.generativeai:generativeai:0.9.0")

val model = GenerativeModel(
    modelName = "gemini-1.5-pro",
    apiKey = BuildConfig.GEMINI_API_KEY,
    generationConfig = generationConfig {
        temperature = 0.7f
        maxOutputTokens = 2048
        topK = 40
        topP = 0.95f
    },
    safetySettings = listOf(
        SafetySetting(HarmCategory.HARASSMENT, BlockThreshold.MEDIUM_AND_ABOVE)
    )
)

On iOS — GoogleGenerativeAI via Swift Package Manager. API identical, difference only in syntax.

For Flutter — google_generative_ai package covers both platforms.

Multimodality: Gemini's Native Advantage

Gemini 1.5 Pro processes text, images, audio, video and PDF in single request with context up to 1 million tokens. For mobile assistant opens scenarios unavailable to other models: pass 30-minute video and request summary, or upload meeting recording for transcription with summary.

Image transmission via Android SDK:

val image = BitmapFactory.decodeResource(resources, R.drawable.photo)
val content = content {
    image(image)
    text("Describe what's happening in this photo")
}
val response = model.generateContent(content)

Files over 20 MB must upload via File API (POST https://generativelanguage.googleapis.com/upload/v1beta/files), not pass inline base64. File API stores file 48 hours, returns file_uri used in subsequent requests.

Streaming and Native Kotlin Coroutines

Gemini Android SDK returns Flow<GenerateContentResponse> for streaming — native Kotlin coroutines integration:

viewModelScope.launch {
    model.generateContentStream(prompt).collect { chunk ->
        val text = chunk.text ?: return@collect
        _uiState.update { it + text }
    }
}

Cleaner than manual SSE parsing. On iOS similar AsyncThrowingStream<GenerateContentResponse, Error>.

Gemini vs Vertex AI: Production Choice

Google offers two paths:

• Google AI (Gemini API) — direct access, simple start, for MVP and small apps
• Vertex AI — enterprise option with extras: fine-tuning, corporate SLA, data not used for training, Google Cloud IAM integration

For mobile app with user data — Vertex AI with server proxy. For prototype or B2B tool without sensitive data — Gemini API sufficient.

Vertex AI SDK on Android: com.google.cloud:google-cloud-aiplatform, but requires service account authentication, implying server layer.

Safety Settings and Content Filtering

Gemini has built-in blocking system by categories: HARASSMENT, HATE_SPEECH, SEXUALLY_EXPLICIT, DANGEROUS_CONTENT. Default threshold BLOCK_MEDIUM_AND_ABOVE — quite aggressive. For medical or legal apps needing sensitive topic discussion, threshold lowered to BLOCK_ONLY_HIGH or BLOCK_NONE for specific categories.

Blocked content response returns finishReason: SAFETY, not HTTP error — must explicitly check this field, otherwise user gets empty response without explanation.

Timeline Estimates

Text assistant with native SDK — 1 week. Multimodal assistant with File API, streaming and server proxy — 3–4 weeks.