An intelligent urban exploration and landmark assistant built with Google Agent Development Kit (ADK), Vertex AI, Google Maps APIs, and Google Cloud services.
Based on the implementation in app/agent.py and app/tools.py, Urban Navigator AI is powered by the following wired Google Cloud services and tool APIs:
- Vertex AI Memory Bank (
PreloadMemoryTool&generate_memories_callback): Automatically persists user preferences, food allergies, and dietary restrictions across sessions. - Google Cloud Firestore (
get_landmarks&add_landmark): Saves and queries landmark records, detour times, ratings, and addresses in thelandmarksdatabase collection. - Google Cloud Storage (GCS) (
generate_landmark_image): Hosts generated high-resolution landmark artwork in a public cloud storage bucket. - Vertex AI Imagen (
gemini-3.1-flash-lite-image): Synthesizes photorealistic vector/watercolor landmark illustrations on demand. - A2UI v0.8 Basic Catalog (
a2ui_callback&A2uiSchemaManager): Renders dynamic rich cards, images, columns, and interactive action buttons natively in the chat interface.
- Google Maps Geocoding API (
geocode_address): Converts addresses and landmark names into latitude and longitude coordinates. - Google Maps Places API (New) (
find_nearby_places): Discovers nearby restaurants, cafes, parks, museums, and points of interest around specified coordinates. - Open-Meteo Live Weather API (
get_live_weather): Retrieves real-time temperature (°F/°C) and wind forecasts for any global city. - OpenStreetMap Nominatim (
search_place_online): Performs real-world online place lookups for landmarks and locations.
┌─────────────────────────────────────────────────────────┐
│ User Web Frontend │
│ (FastAPI Proxy + Custom Glassmorphic UI) │
└────────────────────────────┬────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────┐
│ Urban Navigator ADK Agent │
└──────┬─────────────────────┬─────────────────────┬──────┘
│ │ │
▼ ▼ ▼
┌───────────────┐ ┌───────────────┐ ┌───────────────┐
│ Vertex AI │ │ Cloud │ │ Google Maps │
│ Memory Bank │ │ Firestore │ │ & Places API │
└───────────────┘ └───────────────┘ └───────────────┘
│ │
▼ ▼
┌───────────────┐ ┌───────────────┐
│ Vertex AI │ │ Cloud Storage │
│ Imagen Model │────►│ Public Bucket │
└───────────────┘ └───────────────┘
- Python 3.10+
uvorpip- Google Cloud Platform Project with Firestore, Vertex AI, and Google Maps API enabled.
Create a .env file in the project root:
PROJECT_ID=<your-gcp-project-id>
GOOGLE_MAPS_API_KEY=<your-google-maps-api-key>
GCS_BUCKET_NAME=<your-public-gcs-bucket># Navigate to the app directory
cd app
# Install dependencies
uv sync
# Run local ADK development server
adk web# Navigate to frontend folder
cd frontend
# Set environment variables for local testing
export AGENT_DIRECTORY="app"
export PORT=8080
# Start FastAPI frontend server
uv run python main.py-
Geocode & Nearby Search:
"Geocode Golden Gate Bridge in San Francisco and find nearby cafes"
-
Live Weather & Online Search:
"Get live weather for Tokyo and search for historic landmarks"
-
Database Persistence & AI Illustration:
"Save Eiffel Tower to Firestore and generate a landmark illustration card"
