Building Your First AI-Powered API with Apigee and Vertex AI: A Developer’s Guide

This article walks you through how to create and expose an AI-powered API using Google Cloud’s Vertex AI and Apigee. You’ll build a simple backend in Python that leverages Vertex AI’s PaLM model, then proxy that backend using Apigee to manage, secure, and monitor the API.

Ideal for developers looking to embed generative AI into real-world systems without reinventing the wheel.

Prerequisites

• A GCP project with Vertex AI, Cloud Functions, and Apigee enabled

• IAM permissions for Vertex AI and Cloud Functions

• Basic Python knowledge

• gcloud CLI installed and authenticated
  
  

Step 1: Create a Simple Vertex AI-Powered Function

We’ll use a Cloud Function to connect to the PaLM API and return a generated response.

Install Vertex AI SDK:

pip install google-cloud-aiplatform

Python Code (main.py):

from flask import jsonify

from flask import Request

from vertexai.language_models import TextGenerationModel

import vertexai

def palm_generate(request: Request):

    request_json = request.get_json()

    prompt = request_json.get("prompt", "")

    vertexai.init(project="your-project-id", location="us-central1")

    model = TextGenerationModel.from_pretrained("text-bison")

    response = model.predict(prompt, max_output_tokens=256)

    return jsonify({"completion": response.text})

Reminder: replace your-project-id and us-central1 with your Google Cloud Project ID as well as the targeted Google Cloud Region identifier.

requirements.txt:

Flask==2.2.2

google-cloud-aiplatform

Deploy Cloud Function:

gcloud functions deploy palm-generate \

  --runtime=python310 \

  --trigger-http \

  --allow-unauthenticated \

  --entry-point=palm_generate \

  --region=us-central1

Step 2: Test the Cloud Function

Use curl or Postman:

curl -X POST https://REGION-PROJECT.cloudfunctions.net/palm-generate \

  -H "Content-Type: application/json" \

  -d '{"prompt": "What are three key takeaways from the Agile manifesto?"}'

You should see a text response from PaLM!

Step 3: Set Up Apigee Proxy

Log in to Apigee UI, go to Develop > API Proxies, and click +Proxy.

Proxy Details:

• Proxy Type: Reverse Proxy

• Name: ai-text-generator

• Existing backend: URL of your Cloud Function

• Base Path: /ai-text-generator

Click Next, then Add CORS Policy and enable OAuth2 if needed.

Add API Key Verification (Optional):

Under Policies > PreFlow, insert:

<VerifyAPIKey name="Verify-API-Key"/>

And attach your key validation.  Reminder:  requires creating the API key.

Deploy:

Choose your environment (test/prod), click Deploy, and grab your Apigee endpoint URL.

Step 4: Make a Request to Your Apigee Proxy

curl -X POST https://your-apigee-endpoint/ai-text-generator \

  -H "Content-Type: application/json" \

  -H "x-api-key: YOUR-API-KEY" \

  -d '{"prompt": "Summarize the benefits of serverless architecture"}'

You now have a fully managed, secured, AI-powered API.

Bonus: Add Rate Limiting and Monitoring

In Apigee:

• Spike Arrest Policy:

<SpikeArrest name="RateLimit">

  <Rate>10ps</Rate>

</SpikeArrest>

• Monitoring: Use Apigee’s analytics dashboard to view call volume, latency, and errors.
  
  

Example Architecture Diagram

Client → Apigee Proxy → Cloud Function → Vertex AI → Response

Congratulations! You’ve now built your first AI-powered API using Vertex AI and Apigee. From here, you can expand to:

• Multi-agent orchestration with additional endpoints

• Logging and tracing with Cloud Trace

User access control with Apigee OAuth2