29 - Vertex AI Studio and Gemini API

# 1. Export project environment variables
export PROJECT_ID=$(gcloud config get-value project)
export REGION="us-central1"

# 2. Enable Vertex AI and Cloud Storage APIs
gcloud services enable \
    aiplatform.googleapis.com \
    storage.googleapis.com

# 1. Create a dedicated storage bucket
export BUCKET_NAME="ai-demo-${PROJECT_ID}"
gcloud storage buckets create gs://${BUCKET_NAME} --location=${REGION}

# 2. Download a sample Google Cloud architecture diagram
	curl -s -o sample_diagram.png https://cloud.google.com/static/images/architecture/serverless-three-tier-web-app.png

# 3. Copy the image into your private bucket
gcloud storage cp sample_diagram.png gs://${BUCKET_NAME}/sample_diagram.png

mkdir -p ~/vertex-ai-app
cd ~/vertex-ai-app

cat << 'EOF' > vision_analyzer.py
import os
import sys
import json
import urllib.request
import subprocess

# 1. Initialize Project & Access Tokens
PROJECT_ID = subprocess.check_output("gcloud config get-value project", shell=True).decode().strip()
ACCESS_TOKEN = subprocess.check_output("gcloud auth print-access-token", shell=True).decode().strip()
BUCKET_NAME = os.environ.get("BUCKET_NAME", f"ai-demo-{PROJECT_ID}")

# 2. Determine target image from GCS
if len(sys.argv) > 1:
    filename = sys.argv[1]
    image_uri = f"gs://{BUCKET_NAME}/{filename}" if not filename.startswith("gs://") else filename
else:
    output = subprocess.check_output(f"gcloud storage ls gs://{BUCKET_NAME}/", shell=True).decode()
    images = [line.strip() for line in output.splitlines() if any(line.lower().endswith(ext) for ext in ['.jpg', '.jpeg', '.png', '.webp'])]
    user_images = [img for img in images if 'sample_diagram' not in img]
    image_uri = user_images[-1] if user_images else images[-1]

mime_type = "image/jpeg" if image_uri.lower().endswith(('.jpg', '.jpeg')) else "image/png"

print(f"\n📸 Selected Image: {image_uri}")
print("🤖 Model: Gemini 2.5 Flash")
print("🔍 Analyzing person, scene, and architecture...\n")

# 3. Prompt for Person, Scene & Technical Analysis
prompt = """
Analyze this image in detail:
1. If there is a person/people: Identify if it is a well-known public figure, celebrity, or historical personality. Describe their appearance, clothing, posture, and facial expression.
2. If this is a diagram: Identify the Google Cloud services and request flow depicted.
3. Describe the overall setting, background, and 3 key notable visual details.
"""

payload = {
    "contents": [
        {
            "role": "user",
            "parts": [
                {
                    "fileData": {
                        "mimeType": mime_type,
                        "fileUri": image_uri
                    }
                },
                {
                    "text": prompt
                }
            ]
        }
    ],
    "generationConfig": {
        "temperature": 0.2,
        "maxOutputTokens": 1024
    }
}

# 4. Call Vertex AI Endpoint directly with Gemini 2.5 Flash
url = f"https://us-central1-aiplatform.googleapis.com/v1/projects/{PROJECT_ID}/locations/us-central1/publishers/google/models/gemini-2.5-flash:generateContent"

req = urllib.request.Request(
    url,
    data=json.dumps(payload).encode("utf-8"),
    headers={
        "Authorization": f"Bearer {ACCESS_TOKEN}",
        "Content-Type": "application/json"
    },
    method="POST"
)

with urllib.request.urlopen(req) as resp:
    data = json.loads(resp.read().decode("utf-8"))
    text = data["candidates"][0]["content"]["parts"][0]["text"]
    print("=" * 60)
    print("🤖 GEMINI 2.5 FLASH ANALYSIS RESULT:")
    print("=" * 60)
    print(text)
    print("=" * 60 + "\n")
EOF

nano vision_analyzer.py

cat << 'EOF' > json_extractor.py
import os
import json
import urllib.request
import subprocess

PROJECT_ID = subprocess.check_output("gcloud config get-value project", shell=True).decode().strip()
ACCESS_TOKEN = subprocess.check_output("gcloud auth print-access-token", shell=True).decode().strip()

raw_review = """
Customer Feedback (2026-08-31):
Hi, I ordered 3 pairs of CloudRunner Shoes (Order #94812) on Friday. 
The delivery was lightning fast (2 days!), but the shoe size was slightly smaller than expected.
Rating: 4 out of 5 stars. Customer sentiment: positive but requested an exchange.
"""

prompt = f"""
You are a pure data extraction engine. Extract the following information into a strict JSON object:
- order_id (integer)
- product_name (string)
- quantity (integer)
- rating (integer out of 5)
- sentiment (POSITIVE, NEUTRAL, or NEGATIVE)
- action_required (string)

Return ONLY pure valid JSON, without backticks or markdown formatting.

Text to analyze:
{raw_review}
"""

payload = {
    "contents": [{"role": "user", "parts": [{"text": prompt}]}],
    "generationConfig": {
        "temperature": 0.0,
        "responseMimeType": "application/json"
    }
}

url = f"https://us-central1-aiplatform.googleapis.com/v1/projects/{PROJECT_ID}/locations/us-central1/publishers/google/models/gemini-2.5-flash:generateContent"

req = urllib.request.Request(
    url,
    data=json.dumps(payload).encode("utf-8"),
    headers={
        "Authorization": f"Bearer {ACCESS_TOKEN}",
        "Content-Type": "application/json"
    },
    method="POST"
)

with urllib.request.urlopen(req) as resp:
    data = json.loads(resp.read().decode("utf-8"))
    text = data["candidates"][0]["content"]["parts"][0]["text"]
    
    # Parse and format verified JSON dictionary
    parsed_json = json.loads(text)
    print("\n--- 📊 Extracted Database-Ready JSON ---")
    print(json.dumps(parsed_json, indent=2))
    print("---------------------------------------\n")
EOF

export PROJECT_ID=$(gcloud config get-value project)
export BUCKET_NAME="ai-demo-${PROJECT_ID}"

# 1. Run Multimodal Image Analysis (with Gemini 2.5 Flash)
python3 vision_analyzer.py

# 2. Run Structured JSON Extraction (with Gemini 2.5 Flash)
python3 json_extractor.py

# Delete Cloud Storage bucket and image
gcloud storage rm --recursive gs://${BUCKET_NAME}

# Delete local workspace files
rm -rf ~/vertex-ai-app sample_diagram.png

// Try answering these:
1. What is the defining characteristic of a "Foundation Model" compared to traditional task-specific AI?A) It can only run on quantum computersB) It is a single massive general-purpose base model pre-trained on broad data that can be adapted to hundreds of different tasks via promptingC) It only works for processing numbers in spreadsheetsD) It requires manually labeling 1 million training images for every new question
2. Why would an engineer set the temperature parameter to 0.0 or 0.1 when using Gemini in a software application?A) To cool down the physical server CPUsB) To enforce strict, deterministic, and factual outputs with minimal randomness (ideal for code generation, data extraction, and math)C) To make the model generate fictional storiesD) To delete the model cache
3. What is the primary data privacy benefit of using Google Cloud Vertex AI over free consumer chatbot tools?A) Vertex AI has a dark mode themeB) Vertex AI guarantees enterprise data privacy: your prompts, customer data, and documents are never used to train Google's public foundation modelsC) Consumer chatbots only work during business hoursD) Vertex AI disables all user passwords
4. What does "Multimodality" mean in the Gemini model architecture?A) The model requires 3 separate graphics cards to bootB) The model natively ingests, understands, and reasons across multiple data formats (text, code, images, audio, video) in a single unified systemC) The model only understands English and SpanishD) The model can only generate JPEG images